# Zep — full page content Concatenated body content of every page on www.getzep.com, rendered to Markdown for direct LLM ingestion. Generated as a postbuild step from `.next/server/app/` HTML — see `scripts/generate-llms-full-txt.js`. Each section below is one route. The `**Source:**` URL is the canonical page URL. --- ## Agent memory at enterprise scale — Zep **Source:** https://www.getzep.com/ Memory of users, the business, and work done. Managed, governed, and served at scale. Trusted By AI Teams ## Memory Built on Context Graphs Create memories from any source. Zep constructs the graph. Retrieve relevant, token-efficient context. Any Source Chat History Business Data User Interactions People, things, and [how they change](#memory-validity). Automated Context Assembly context\_response.py ``` Emily prefers cycling to jogging (Valid: 2024-11-14 — present) # Observations Emily's blood pressure ranges 5% lower after cycling workouts. ``` Context Lake 1,402,891 active graphs USER user\_8a32e1f9 247 1,204 2m ORG customer\_acme\_co 89 412 14s AGENT agent\_voyager 1,820 9,330 now DOMAIN domain\_billing 64 218 5m USER user\_d72b40c1 186 731 1m ORG customer\_initech 142 603 38s AGENT agent\_atlas\_v3 2,447 10,580 12s DOMAIN domain\_pricing 31 127 8m USER user\_4cc1ee07 312 1,486 4m ORG customer\_pied\_piper 77 298 22s AGENT agent\_helios 1,108 5,242 45s DOMAIN domain\_routing 48 174 3m USER user\_91ab7d2e 208 942 18s ORG customer\_globex 104 481 2m AGENT agent\_orion\_v2 1,640 7,820 8s DOMAIN domain\_inventory 56 198 6m Access control Retention Provenance Audit ENTERPRISE SCALE MEMORY ## Introducing the Context Lake Millions of context graphs, governed and served as one system. The data-lake pattern, applied to agent context. Powered by [Konig](/platform/agent-knowledge-graph/), Zep’s proprietary graph database service. [Learn about the Context Lake](/platform/context-lake/) Retrieval latency · p95 10K 148ms 100K 152ms 1M 156ms 10M 161ms 100M 168ms **Graph Size** ### Sub-200ms retrieval Retrieval stays under 200 milliseconds, regardless of graph size or count. ### Governed at the data layer Authorization, retention, and audit live in the data layer, not bolted on. Policy applies across every graph, every query, every layer. [Learn more](#governance) Robbie strongly favors Adidas shoes. traced\_from Chat message user\_8a32e1f9 2024-09-07 > “I only wear Adidas shoes. I love them!” ### Provenance preserving Every fact in the graph traces back to the source episode that produced it. Audit any answer back to where it came from. Recognition ## Agent memory _infrastructure_ for the enterprise. [Read the report](/analysts/sp-market-intelligence-report/) [ S&P Global Market Intelligence ### Zep tackles agent memory limitations through its temporal context graph. S&P Global Market Intelligence · April 2026 ](/analysts/sp-market-intelligence-report/) We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack. — Melissa Incera, S&P Global Market Intelligence ## Memory that understands when things change When new information contradicts what’s in the graph, Zep invalidates the old fact. Your agent reasons with the latest decisions, traits, and behaviors. Old facts stay as history. Ask what’s true now, or what was true on any past date. R Robbie 2024-09-07 · 14:27 I only wear Adidas shoes. I love them! Facts - Robbie only wears Adidas shoes. - Robbie strongly favors Adidas shoes. soleworks.com /account/returns/SO-48219 Soleworks Return · Order #SO-48219 · Adidas Ultraboost 22 Reason for return Product fell apart Additional comments These Adidas fell apart after three weeks and I'm furious . I'll be buying Nike from now on. Facts - Robbie only wears Adidas shoes. - Robbie strongly favors Adidas shoes. - Robbie ’s Adidas shoes fell apart . - Robbie is returning their Adidas shoes. - Robbie is angry about their Adidas shoes. - Robbie intends to wear Nike shoes. ## Memory patterns become Observations Zep analyses the structure of the graph to surface _Observations_: patterns, recurrences, and co-occurrences in memory. Your agent gains a global perspective, beyond facts and summaries. [Learn about agent memory](/product/agent-memory/) Jane has upgraded within _two weeks_ of each of the last _three product launches_. Apr 12, 2025 Jane upgraded to Pro v3 . +9d after launch Aug 4, 2025 Jane upgraded to Pro v4 . +11d after launch Nov 19, 2025 Jane upgraded to Pro v5 . +6d after launch ## Three Lines of Code Add memory to your agent in minutes. Works with any agent framework, or none. [Read the Quickstart](https://help.getzep.com/v3/quick-start-guide) quickstart.py ```python # Add messages and get context in one call response = client.thread.add_messages( thread_id=thread_id, messages=[Message(name="Jane" , role="user" , content="I'd like to upgrade my plan..." )], return_context=True , )   # Add business data to the user's graph client.graph.add( user_id=user_id, type ="json" , data=json.dumps({"event" : "plan_upgrade" , "to" : "pro" , "mrr" : 49 }), )   # Get relevant context user_context = client.thread.get_user_context(thread_id=thread_id) ``` ## Governed at the data layer Govern context across thousands of agents, users, and context sources. Access control ### Attribute-based access control Control what context agents can access and what they can do with it. Retention ### Retention policies Retention is policy-driven. Data expires on the schedule you set. Legal hold blocks deletion when compliance requires it. Audit ### Audit and API logs Detailed logs of every request and policy decision, ready for audit. ## More accurate. Faster. Fewer tokens. Agent memory systems often trade one for another. [Zep leads on all three.](/research/) LoCoMo 94.7 % accuracy Retrieval latency 155 ms Context size 5,760 tokens LongMemEval 90.2 % accuracy Retrieval latency 162 ms Context size 4,408 tokens [See the methodology and full results](/research/) ## Built-in observability Latency, error rate, retrieval activity, and ingest throughput across every project. ## Analytics Usage, latency, and reliability for your account. Episodes Added 1.4M Episodes Processing 847 Graphs Processing 12 Graphs Created 24.7K Users Created 18.3K Avg Latency 187 ms Error Rate 0.41% ### Graphs Created New graphs over time. ### Users Created New users over time. ### Episodes Added New episodes over time. ### Retrieval Activity Context retrieval and graph search requests. Context Retrievals Graph Searches ## Choose your deployment model The trust boundary moves with your deployment. Choose where compute, data, and keys live. [Learn more.](/enterprise/) Trust boundary · Zep Zep Cloud Compute Data Keys Managed ### Cloud Zep's managed service. No infrastructure to run. Start in minutes. - SOC 2 Type II - HIPAA BAA Trust boundary · split Zep Cloud Compute Data Your KMS Keys AWS · GCP · Azure BYOK ### Cloud + Your Own Keys Zep's managed service with your own encryption keys. You control the keys; data at rest is encrypted with them. - SOC 2 Type II - HIPAA BAA Trust boundary · You Your VPC Zep service Compute Data Keys BYOC ### Bring Your Own Cloud Zep deployed inside your VPC. Your network, your perimeter, your compliance boundary. [Security & compliance](https://trust.getzep.com) ## What teams are saying Voices from the teams running Zep in production. Zep is one of the most exciting things I've seen for real-world agent use cases in a long time. Ken Collins VP of Product, Torq and GenAI Expert [](https://www.linkedin.com/in/metaskills/) Unlike other systems that only retrieve static documents, Zep uses a temporal knowledge graph to combine conversations and structured business data, keeping track of how things change over time. Lior Sinclair Founder/CEO, AlphaSignal [](https://www.linkedin.com/feed/update/urn:li:activity:7287958207971434496) Zep AI was instrumental in enabling the Sidekick's personalized experience through dynamic memory retrieval. Mark Losey CTO at Flockx [](https://www.linkedin.com/in/markalosey/) By organizing memories into structured episodes and extracting key insights, it builds smarter, more intuitive AI agents that revolutionize how businesses harness intelligence. Vijay Morampudi Senior Director - AI CoE, Axtria [](https://www.linkedin.com/in/vijaymorampudi/?originalSubdomain=in) Zep is one of the most exciting things I've seen for real-world agent use cases in a long time. Ken Collins VP of Product, Torq and GenAI Expert [](https://www.linkedin.com/in/metaskills/) Unlike other systems that only retrieve static documents, Zep uses a temporal knowledge graph to combine conversations and structured business data, keeping track of how things change over time. Lior Sinclair Founder/CEO, AlphaSignal [](https://www.linkedin.com/feed/update/urn:li:activity:7287958207971434496) Zep AI was instrumental in enabling the Sidekick's personalized experience through dynamic memory retrieval. Mark Losey CTO at Flockx [](https://www.linkedin.com/in/markalosey/) By organizing memories into structured episodes and extracting key insights, it builds smarter, more intuitive AI agents that revolutionize how businesses harness intelligence. Vijay Morampudi Senior Director - AI CoE, Axtria [](https://www.linkedin.com/in/vijaymorampudi/?originalSubdomain=in) ## Start shipping reliable, personalized agents ## Latest from the Zep blog [View all posts](https://blog.getzep.com) Loading latest articles… --- ## About — Zep **Source:** https://www.getzep.com/about/ AGENT Production agent across deal-flow, portfolio review, and weekly insights. Entities 1,782 +42 in last 24h Facts 9,156 +188 in last 24h Episodes 2,449 Last 7d Last updated 3s ago Live ingestion All --- ## AI Agents Guides: Memory, Context & Evaluation **Source:** https://www.getzep.com/ai-agents/ [**We're hiring!** Come build with us →](/careers/) ## Key takeaways - **Agent memory is the category** — the persistent knowledge an agent carries across sessions and sources. Chat buffers, vector stores, and markdown files are all partial attempts to implement it. - **A [temporal context graph](/ai-agents/temporal-knowledge-graph/) is the durable structure** for it: facts with provenance and a validity window, so the agent reasons over what's true now versus what was true then. - **RAG and agent memory are complementary, not the same.** RAG retrieves static documents by similarity; [agent memory](/ai-agents/what-is-agent-memory/) tracks evolving facts about users and the business over time. Most production agents use both. - **A [Context Lake](/platform/context-lake/) is the infrastructure that implements agent memory at enterprise scale** — a governed system of context graphs that manages, governs, and serves what agents need to know, with sub-200ms retrieval. ## Start here If you're new to the topic, read [What is agent memory?](/ai-agents/what-is-agent-memory/) first — it defines the category and explains why chat history and RAG don't scale to it. From there, [Agent memory vs RAG](/ai-agents/agent-memory-vs-rag/) draws the distinction most teams get wrong, and [What is a temporal knowledge graph?](/ai-agents/temporal-knowledge-graph/) explains the structure underneath. When you're ready to build, [How to give an AI agent long-term memory](/ai-agents/how-to-give-ai-agents-long-term-memory/) walks through the approaches, and the [persistent-memory tutorial](/ai-agents/persistent-memory-for-ai-agents/) is the hands-on version. The infrastructure these guides point to is the [Context Lake](/platform/context-lake/) — Zep manages agent memory (ingest, construct, invalidate, evolve), governs it (ABAC, retention, audit), and serves it as assembled context with sub-200ms p95 retrieval. The graph itself is built with [Graphiti](/platform/graphiti/), Zep's open-source temporal context graph library, which runs on [Konig](/platform/agent-knowledge-graph/), Zep's proprietary graph database service, at scale. For how Zep's accuracy is measured, see the [LoCoMo and LongMemEval results](/research/). ## Core concepts - [What is agent memory? The category, defined: what an agent knows over time about users, the business, and the work, and why chat history and RAG fall short. Read guide](/ai-agents/what-is-agent-memory/) - [What is a Context Lake? The infrastructure that implements agent memory at enterprise scale, and the data-lake parallel explained. Read guide](/ai-agents/what-is-a-context-lake/) - [Agent memory vs RAG Where RAG breaks for agents, what agent memory adds, and how to use both together. Read guide](/ai-agents/agent-memory-vs-rag/) - [What is a temporal knowledge graph? Standard versus temporal graphs, what bi-temporal means, and why agents need validity and provenance. Read guide](/ai-agents/temporal-knowledge-graph/) ## How-to & tutorials - [How to give an AI agent long-term memory The approaches compared (full history, summarization, RAG, temporal context graph), with code. Read guide](/ai-agents/how-to-give-ai-agents-long-term-memory/) - [Persistent memory for AI agents (tutorial) A hands-on build: ingest sources, construct the graph, retrieve relevant context per turn. Read guide](/ai-agents/persistent-memory-for-ai-agents/) - [How to reduce LLM hallucinations Why models hallucinate, how often the frontier does it on AA-Omniscience, and the layered fix — grounding, abstention, verification, and agent memory. Read guide](/ai-agents/reducing-llm-hallucinations/) ## Testing & benchmarks - [How to test agent memory Measure context completeness first, then answer correctness, latency, and token use, across sessions and over time. Read guide](/ai-agents/how-to-test-agent-memory/) - [LLM evaluation framework Evaluate agent applications on your own data, plus the memory benchmarks (LoCoMo, LongMemEval). Read guide](/ai-agents/llm-evaluation-framework/) ## Frequently asked questions ### What is agent memory? Agent memory is everything an AI agent knows across time about the users, the business, and the world it operates in — so it can reason, personalize, and act without starting from scratch every turn. It's the category; chat buffers, vector stores, and Context Lakes are different ways to implement it. See [What is agent memory?](/ai-agents/what-is-agent-memory/) ### How is agent memory different from RAG? RAG retrieves static documents by similarity at query time. Agent memory tracks evolving, provenance-stamped facts about users and the business over time, with a sense of what's true now versus what was true then. They're complementary — most production agents use RAG for documents and agent memory for state. See [Agent memory vs RAG](/ai-agents/agent-memory-vs-rag/). ### How do you give an AI agent long-term memory? Add a memory layer that builds a [temporal context graph](/ai-agents/temporal-knowledge-graph/) from the agent's inputs and serves the relevant context back per turn — rather than stuffing chat history into the context window. With Zep this is a few lines of code and works with any agent framework. See [How to give an AI agent long-term memory](/ai-agents/how-to-give-ai-agents-long-term-memory/). ### How do you test agent memory? Measure context completeness first — did the system retrieve the facts needed to answer — then answer correctness, retrieval latency, and token use, across multiple sessions and over time. Industry benchmarks include LoCoMo and LongMemEval. See [How to test agent memory](/ai-agents/how-to-test-agent-memory/). ### What's the best way to do agent memory at enterprise scale? A [Context Lake](/platform/context-lake/) — a governed system of context graphs that manages, governs, and serves agent memory across millions of users with sub-200ms retrieval, attribute-based access control, retention, and audit. Zep is the Context Lake for AI agents. * * * _Part of the Zep AI agent memory guides. Built on [Graphiti](/platform/graphiti/) and [Konig](/platform/agent-knowledge-graph/). See the [research and benchmarks](/research/)._ --- ## S&P Global Market Intelligence on Zep **Source:** https://www.getzep.com/analysts/sp-market-intelligence-report/ Industry analyst coverage > “We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack.” — **Melissa Incera**, S&P Global Market Intelligence · April 2026 22 pages · PDF ## Get the report A look inside ## What’s in the report. 01 ### The category Where agent memory sits in the enterprise stack, and how it's emerging as a distinct category of infrastructure. 02 ### The architecture Zep's temporal context graph, and what makes it different from chat-history memory and vector-store retrieval. 03 ### The deployments How enterprise teams are putting Zep into production today. Trusted By AI Teams --- ## Zep vs. AWS AgentCore Memory: Neutral Alternative **Source:** https://www.getzep.com/aws-agentcore-memory-alternative/ Zep vs. AWS AgentCore Memory AWS Bedrock AgentCore Memory is a managed memory service for agents inside the AWS ecosystem. Zep is a neutral, multi-LLM, multi-cloud Context Lake that manages, governs, and serves agent memory on temporal context graphs. Key takeaways ## A hyperscaler primitive, or a _neutral_ memory layer - AWS AgentCore Memory ([docs](https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/memory.html)) is bound to the AWS/Bedrock ecosystem; Zep is a neutral, multi-LLM, multi-cloud [Context Lake](/platform/context-lake/). - The decision is lock-in vs. neutrality — S&P Global Market Intelligence named the neutral, multi-LLM memory layer as Zep's opportunity ([coverage](https://www.getzep.com/research/sp-global-market-intelligence-zep-coverage/)). - Zep runs managed, BYOK, or BYOC on AWS/GCP/Azure, builds bi-temporal context graphs, and reports 94.7% LoCoMo accuracy at sub-200ms ([results](https://www.getzep.com/research/)). The distinction ## An AWS primitive vs. a neutral layer **What AWS AgentCore Memory is.** AgentCore Memory is part of Amazon Bedrock AgentCore — AWS's managed building blocks for agents. It provides short- and long-term memory for agents running in the AWS/Bedrock environment, integrated with the rest of the AWS agent stack. For teams all-in on AWS, that integration is the appeal. **What Zep is.** Zep is a dedicated, neutral memory layer — the [Context Lake](/platform/context-lake/) for AI agents. It builds bi-temporal context graphs from chat, business data, and documents (via open-source [Graphiti](/platform/graphiti/) on Konig, Zep's proprietary graph database service), serves token-efficient context in sub-200ms p95, and runs as managed cloud, with your own keys (BYOK), or inside your VPC (BYOC) on the cloud you choose. It's model- and framework-agnostic by design. ### Agent Runtime LangChain · LlamaIndex · CrewAI · Google ADK · custom Any agent framework — or none. The Context Lake is invoked through a single SDK. ### Ingestion chat · JSON · documents · app events Raw signal arrives from any source the agent touches. ### Context Assembly context blocks · templates · token-efficient Relevant context is assembled on demand into token-efficient blocks. ### Graphiti entity extraction · relationships · ontology · invalidation Signal becomes a temporal context graph as new facts arrive and stale ones are invalidated. ### Retrieval sub-200ms · auto-optimized · provenance-linked · policy-filtered Selects what's relevant and what adds the most information within the token budget. ### Governance ABAC · multi-tenant isolation · customer key encryption · retention policies · audit · provenance Native to the data layer, not a layer bolted on. Every read and write is policy-gated for access and provenance; retention runs across the data lifecycle. ### _Konig_ entities · facts & edges · decision traces · episodes Temporal context graph with provenance — sub-200ms retrieval at scale. How they compare ## AgentCore Memory vs. Zep, side by side AWS AgentCore Memory Zep Ecosystem Bound to AWS / Bedrock Neutral — any model, any cloud Model providers AWS-centric OpenAI, Anthropic, Meta, others Memory model Managed short-term (session events) + long-term (async-extracted insights), semantic retrieval Bi-temporal context graph (provenance + validity) Temporal reasoning No — extraction-based; no temporal graph “What's true now / what was true then,” auto fact invalidation Deployment AWS Managed, BYOK, or BYOC (AWS/GCP/Azure) Benchmarks — 94.7% LoCoMo (155ms), 90.2% LongMemEval (162ms) Lock-in risk Higher (ecosystem-bound) Lower (portable across stacks) The strategic question ## Lock-in vs. _neutrality_ S&P Global Market Intelligence (451 Research) named this directly: Zep's opportunity is to be the neutral, multi-LLM memory layer for enterprises wary of hyperscaler lock-in — one consistent context strategy across model providers and clouds. Hyperscaler memory primitives are convenient if you're committed to that ecosystem and using “good enough” memory bundled in. The risk is that your agents' memory — among the most valuable, sticky data you have — becomes bound to one vendor's stack. When to choose ## Pick the tool that fits the strategy Choose AgentCore Memory when You're fully committed to AWS/Bedrock and the bundled primitive meets your needs. - Fully committed to AWS/Bedrock - You want the tightest native AWS integration - Memory needs are well served by the bundled primitive Choose Zep when You want to avoid lock-in and keep a consistent memory layer across models and clouds. - Neutrality across models and clouds - Temporal, provenance-tracked, governed memory (ABAC, retention, audit) - Regulated workloads with BYOK/BYOC deployment control - Memory quality at scale as a first-class requirement Get started ## Keep your agent memory _portable_ FAQ ## Frequently asked questions ### Is AWS AgentCore Memory enough for enterprise agent memory? If you're all-in on AWS and need basic managed memory, it can be. If you need neutrality across models/clouds, temporal reasoning, provenance, and portable governance, evaluate a dedicated layer like Zep. ### Can Zep run on AWS? Yes — managed, with your own keys, or inside your own VPC on AWS (or GCP/Azure). You keep deployment and key control without ecosystem lock-in. ### Does Zep work with Amazon Bedrock models? Zep is model-agnostic and works across providers including those on Bedrock, as well as OpenAI, Anthropic, and others. --- ## Careers — Zep **Source:** https://www.getzep.com/careers/ [**We're hiring!** Come build with us →](/careers/) Careers Join a high-agency team building the Context Lake for AI agents. What we're building ## Memory across every user, every domain, every _agent_. Agents are limited by the context available to them. Most memory systems see only the conversation in front of them — not the user, the business, or the work that came before. Zep changes that. We build the infrastructure that lets agents remember and reason across every source they touch: chat, documents, events, business data. [The _Context Lake_](/platform/context-lake/) is governed and temporal, and scales to millions of context graphs per deployment. Backed by leading investors - - - Plus angels at industry-leading companies including Vercel, Google, and Airtable. ## How _we work_ We're a small, distributed team that works closely together. We pair on hard problems, review each other's designs, and treat learning as part of the job rather than something that happens after hours. We ask a lot of questions: of customers, of teammates, of our own assumptions. When we find pain, we go fix it. We expect the same back: ask questions early, push back when you disagree, and care about the people on the other end of the API. ## Open _roles_ Open positions from Work at a Startup. ### Roles are temporarily unavailable here The public Work at a Startup feed could not be loaded during this request. Open the YC jobs board directly for the latest roles. Benefits ## What _we offer_. - ### Great healthcare Platinum medical, dental, and vision insurance. - ### Compensation Competitive salary and equity. 401K plan with employer matching. Unlimited PTO. - ### Flexible WFH Flexible in-office culture in San Francisco. Remote options with periodic travel for team members outside the Bay Area. - ### Cell phone stipend Monthly stipend toward your mobile plan. Our process ## How _we hire_. At Zep, we move quickly when we spot talent. 1. ### Introductory video call A short call with Daniel, our founder. 2. ### Team interview How well you fit our collaborative, team-focused environment. 3. ### Final interview with our CEO A one-on-one on your role, goals, and contributions to Zep's growth. ## Come build _with us_. Join a high-agency team building the Context Lake for AI agents. --- ## Cognee Alternative — Agent Memory That Runs at Enterprise Scale **Source:** https://www.getzep.com/cognee-alternative/ Zep vs. Cognee Cognee is an open-source toolkit you wire together and operate. Zep is agent memory at enterprise scale, delivered as a managed Context Lake — bi-temporal context graphs, governance in the data layer, and sub-200ms retrieval, out of the box. Key takeaways ## The pieces, or the _system_ - Cognee is an open-source ECL pipeline you point at your own graph and vector backends, then host and operate. Zep is a managed [Context Lake](/platform/context-lake/) — one runtime, one SDK. - Zep models time at the data layer: every fact carries valid-from and valid-to timestamps, so point-in-time queries follow from the model rather than logic you build above it. - Governance — ABAC, retention with legal hold, audit — is enforced in the data layer, alongside SOC 2 Type II, HIPAA, and BYOC. - Zep reports 94.7% LoCoMo accuracy at 87ms p50 and 90.2% on LongMemEval at 104ms ([results](/research/)). The distinction ## Cognee gives you the pieces. Zep runs the system. **What Cognee is.** Cognee is an open-source ECL (extract, cognify, load) pipeline. You point it at your own graph and vector backends, then host and operate the result yourself. For teams that want a fully open-source core and maximum control over those backends, that's the appeal. **What Zep is.** Zep manages, governs, and serves agent memory for you — the [Context Lake](/platform/context-lake/) for AI agents. It ingests chat, JSON, app events, documents, and business data through a single SDK, unifies them in one bi-temporal context graph per subject (via open-source [Graphiti](/platform/graphiti/) on [Konig](/platform/agent-knowledge-graph/), Zep's proprietary graph database service), and serves token-efficient context in sub-200ms — with no backend cluster to size, shard, and keep alive. ### Agent Runtime LangChain · LlamaIndex · CrewAI · Google ADK · custom Any agent framework — or none. The Context Lake is invoked through a single SDK. ### Ingestion chat · JSON · documents · app events Raw signal arrives from any source the agent touches. ### Context Assembly context blocks · templates · token-efficient Relevant context is assembled on demand into token-efficient blocks. ### Graphiti entity extraction · relationships · ontology · invalidation Signal becomes a temporal context graph as new facts arrive and stale ones are invalidated. ### Retrieval sub-200ms · auto-optimized · provenance-linked · policy-filtered Selects what's relevant and what adds the most information within the token budget. ### Governance ABAC · multi-tenant isolation · customer key encryption · retention policies · audit · provenance Native to the data layer, not a layer bolted on. Every read and write is policy-gated for access and provenance; retention runs across the data lifecycle. ### _Konig_ entities · facts & edges · decision traces · episodes Temporal context graph with provenance — sub-200ms retrieval at scale. Benchmarks ## Accuracy, latency, and token efficiency — published Zep's results on the two standard long-running memory benchmarks, single retrieval call, no agentic loops. - **LoCoMo · 1,540 questions** — 94.7% accuracy, 87ms retrieval at p50, 5,760 tokens per query. - **LongMemEval · 500 questions** — 90.2% accuracy, 104ms retrieval at p50, 4,408 tokens per query. - Cognee publishes its own evaluation figure, which is measured on a different task and is not directly comparable to LoCoMo. See the full [methodology and results](/research/). How they compare ## Cognee vs. Zep, side by side Cognee Zep Delivery Open-source ECL toolkit — self-hosted and operated Managed Context Lake — one runtime, one SDK Data sources Point at your own graph + vector backends Chat, JSON, events, documents, business data via one SDK, unified per subject Entities & schema Auto-generated ontologies you then correct Custom entities and edges, your schema enforced at ingest Temporal model Not bi-temporal at the data model Bi-temporal facts (valid-from / valid-to), point-in-time queries Governance Assemble it yourself ABAC, retention with legal hold, audit — in the data layer Deployment Self-host the stack Managed, BYOK, or BYOC (AWS / GCP / Azure) Benchmarks Own eval (not comparable to LoCoMo) 94.7% LoCoMo (87ms), 90.2% LongMemEval (104ms) Scale You size and operate the stack Millions of context graphs, sub-200ms at scale When to choose ## Pick the tool that fits the team Stay with Cognee when You want a fully open-source core and you're prepared to assemble and operate the stack. - You want a fully open-source core and maximum control over graph and vector backends - Assembling and operating your own memory stack is acceptable — or preferred - A single-developer or early-stage project matters more than governed scale Choose Zep when you need Agent memory served as a managed runtime, not a stack you host and operate. - Bi-temporal facts with point-in-time queries built into the data model - Business data integrated alongside chat — CRM, support, billing, events - Custom entities and relationships, with your schema enforced - Retrieval that holds at sub-200ms across millions of subjects - Entity-level governance — ABAC, retention, audit — plus SOC 2 Type II, HIPAA, BYOC Get started ## Ready to run agent memory in _production_? FAQ ## Frequently asked questions ### Is Cognee a good agent-memory option? Cognee is an open-source ECL (extract, cognify, load) pipeline you point at your own graph and vector backends, then host and operate. If you want a fully open-source core and maximum control over those backends, it's a fit. If you need agent memory served as a managed runtime — with bi-temporal facts, entity-level governance, and sub-200ms retrieval at scale — evaluate Zep. ### Does Zep replace the graph and vector stores Cognee assembles? Yes. Zep is a managed Context Lake — one runtime and one SDK. The graph, vector, and BM25 indexes are held and served for you, so there is no cluster of backends to size, shard, and keep alive. ### Can Zep run inside my own environment? Zep runs managed, with your own keys (BYOK), or fully inside your VPC (BYOC) on AWS, GCP, or Azure. The trust boundary moves with the deployment. --- ## Zep for Emerging Companies — enterprise agent memory, $13,000 year one **Source:** https://www.getzep.com/emerging/ Zep for Emerging Companies $13,000 for your first year, with more than $40,000 in total discounts over the three-year term. SOC 2 Type II, HIPAA BAA, guaranteed rate limits, and BYOK, for companies that have raised between $1M and $10M. ## Who it’s _for_ 01 You've raised between $1M and $10M. 02 You're building an agent, copilot, or AI-native app that gets better the more it remembers. Memory itself isn't your product. 03 You're moving to production, and your customers are starting to ask about SOC 2, HIPAA, and data handling. 04 You're on Free, Flex, or Flex Plus, and a customer's security review is asking for paperwork the self-serve plans don't include. ## What’s _included_ Everything in Flex Plus, including Smart Context Assembly, Observations, 20 custom entity and edge types, webhooks, and the MCP Server for up to 50 users. The program adds the controls your buyers require: SOC 2 Type II reporting HIPAA BAA and DPA Guaranteed rate limits Audit and API logs, retained for one year BYOK via AWS KMS Unlimited projects EU data residency on request Implementation fee waived Shared Slack channel with the Zep team; onboarding led by a forward-deployed engineer More than double the included usage of Flex Plus, pooled for the year with no monthly expiry BYOC, a dedicated account manager, and custom legal terms are available on standard Enterprise agreements only. The program uses Zep’s standard terms and a fixed order form with zero redlines. ## No renewal _surprises_ Your year-two and year-three prices are on the order form before you sign. The discount steps down on a fixed schedule over three years and is locked in writing the day you start. Year one $13,000 Years Two and Three Prices are locked with discounts and revealed on the order form More than $40,000 in total discounts over the three-year term. ## From application to _signature_ We review every application. Qualified companies get a 30-minute fit call covering the full three-year schedule and usage rates, and the order form follows: one fixed document, priced exactly as discussed, no redlines. With no negotiation cycle to run, most companies are up and running within days of the call. The program is a 12-month commitment, prepaid annually. ## What we ask in _return_ Program pricing carries a few commitments. - Regular feedback on new features. - Participation in quarterly product reviews, sometimes alongside other program companies. - A case study together within the first 12 months. ## _Eligibility_ Eligibility is measured in dollars raised, not round names. Companies on SAFEs or without a named round qualify. $1M–$10M in total funding raised (we verify) Product companies building for many customers — not agencies or consultancies New to Zep, or on Free / Flex / Flex Plus today (not for accounts already on Enterprise) One program term per company, including subsidiaries and affiliates. A company-domain email address Raised more than $10M? The program band ends at $10M. Submit anyway and we’ll route you to a standard Enterprise conversation. Raised less than $1M? Free and Flex are self-serve and cover most teams at that stage. The program opens to you once you cross $1M. Apply ## Check your _eligibility_ Tell us about your company and what you’re building. We verify funding at application, then follow up with next steps on your $13,000 first year. ### Apply to the program ## _FAQ_ What happens after year three? You renew at the full program price, or move to a standard Enterprise agreement priced to your workload. How is funding verified? We check total funding raised against Crunchbase or PitchBook. Eligibility is measured once, at application — a later raise doesn't change your locked schedule. Can existing Flex or Flex Plus customers switch? Yes. Any unused prepaid time credits against your program invoice when you convert. What do years two and three cost? Both prices are printed on the order form before you sign, and neither changes after signature. The discount steps down on a fixed schedule. We walk through the full three-year numbers on the fit call. What happens if we use more than the included credits? Usage past the annual pool is billed monthly at a flat per-credit rate that stays the same in every program year. We disclose the rate on the fit call. You'll see alerts at 80% and 100% of your pool, plus a pace alert if your run rate projects early exhaustion. Can we negotiate the agreement? No. The program runs on Zep's standard agreement and a fixed order form. That constraint is why the timeline is days instead of months: no redlines, no legal cycles, the same terms for every participant. What does the program ask of us? Regular feedback on new features, participation in quarterly product reviews, logo usage, and a case study within your first 12 months. All of it is written into the order form. How do payments work? The program is a 12-month commitment, prepaid annually by ACH or card. Overage, if any, is billed monthly. Payment terms outside annual prepay are considered case by case. Can we move to standard Enterprise mid-term? Yes, at any time. Unused prepaid amounts credit pro rata toward the new agreement. Do SAFEs count toward the funding requirement? Yes. Eligibility is measured in dollars raised, not round names, so companies on SAFEs or without a named round qualify. Is EU data residency available? Yes. The program runs on Zep Cloud in the US by default, with EU data residency available on request. Note the requirement in your application. Not sure where you land? [Check your eligibility](#apply) --- ## Enterprise — Zep **Source:** https://www.getzep.com/enterprise/ Enterprise Production-grade agent memory. Deploy in Zep’s cloud, your VPC, or with your own keys. [See the benchmarks](/research/). Trusted By AI Teams ## Choose your deployment model The trust boundary moves with your deployment. Choose where compute, data, and keys live. [Learn more.](/enterprise/) Trust boundary · Zep Zep Cloud Compute Data Keys Managed ### Cloud Zep's managed service. No infrastructure to run. Start in minutes. - SOC 2 Type II - HIPAA BAA Trust boundary · split Zep Cloud Compute Data Your KMS Keys AWS · GCP · Azure BYOK ### Cloud + Your Own Keys Zep's managed service with your own encryption keys. You control the keys; data at rest is encrypted with them. - SOC 2 Type II - HIPAA BAA Trust boundary · You Your VPC Zep service Compute Data Keys BYOC ### Bring Your Own Cloud Zep deployed inside your VPC. Your network, your perimeter, your compliance boundary. [Security & compliance](https://trust.getzep.com) ## SOC 2 Type II and HIPAA Zep is SOC 2 Type II certified and offers HIPAA Business Associate Agreements. Compliance posture, attestation reports, and security documentation are available in the Trust Center. [](https://trust.getzep.com)[](https://trust.getzep.com) ## Governed at the data layer Govern context across thousands of agents, users, and context sources. Access control ### Attribute-based access control Control what context agents can access and what they can do with it. Retention ### Retention policies Retention is policy-driven. Data expires on the schedule you set. Legal hold blocks deletion when compliance requires it. Audit ### Audit and API logs Detailed logs of every request and policy decision, ready for audit. Recognition ## Agent memory _infrastructure_ for the enterprise. [Read the report](/analysts/sp-market-intelligence-report/) [ S&P Global Market Intelligence ### Zep tackles agent memory limitations through its temporal context graph. S&P Global Market Intelligence · April 2026 ](/analysts/sp-market-intelligence-report/) We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack. — Melissa Incera, S&P Global Market Intelligence ## Talk to the team --- ## Zep vs. HydraDB: A HydraDB Alternative **Source:** https://www.getzep.com/hydradb-alternative/ Zep vs. HydraDB HydraDB and Zep take a similar architectural position — a temporal context graph that fuses graph traversal, vector search, and time, instead of a flat vector index. Zep is the Context Lake for AI agents, an established managed platform; HydraDB is a newer context-graph memory database. Key takeaways ## Same approach, different _maturity_ - Zep and HydraDB share an approach: a **temporal context graph** (graph + vector + time) as the memory layer for agents, rather than a flat vector index. - They differ on maturity and operations: Zep is an established, managed [Context Lake](/platform/context-lake/) with published benchmarks, enterprise governance, proven scale, and analyst validation; HydraDB is a newer entrant with a smaller public track record. - If you want a managed, governed [agent-memory](/ai-agents/what-is-agent-memory/) platform you can deploy in production today — with SOC 2 Type II, HIPAA, and BYOK/BYOC — Zep is the lower-risk choice. The distinction ## Two temporal context graphs, two track records **What HydraDB is.** HydraDB ([hydradb.com](https://hydradb.com/)) provides infrastructure to build and scale an agent's context and memory layer. It describes a composite data layer that fuses a “Git-style temporal graph” for relational integrity with a high-dimensional vector data layer for semantic breadth, and frames recall as closer to a “personalized PageRank for memory” than vector similarity. It targets persistent, cross-session context for customer-support bots, research copilots, and internal knowledge assistants. **What Zep is.** Zep is the [Context Lake](/platform/context-lake/) for AI agents — a managed platform that builds **bi-temporal** context graphs (via the open-source [Graphiti](/platform/graphiti/)), where every fact carries a validity window and provenance. It serves millions of graphs at sub-200ms p95, governs memory in the data layer (ABAC, retention, audit), and deploys managed, BYOK, or BYOC. It reports 94.7% on LoCoMo and 90.2% on LongMemEval ([results](/research/)), with the architecture documented in the [Zep paper](https://arxiv.org/abs/2501.13956) and external validation from S&P Global Market Intelligence. ### Agent Runtime LangChain · LlamaIndex · CrewAI · Google ADK · custom Any agent framework — or none. The Context Lake is invoked through a single SDK. ### Ingestion chat · JSON · documents · app events Raw signal arrives from any source the agent touches. ### Context Assembly context blocks · templates · token-efficient Relevant context is assembled on demand into token-efficient blocks. ### Graphiti entity extraction · relationships · ontology · invalidation Signal becomes a temporal context graph as new facts arrive and stale ones are invalidated. ### Retrieval sub-200ms · auto-optimized · provenance-linked · policy-filtered Selects what's relevant and what adds the most information within the token budget. ### Governance ABAC · multi-tenant isolation · customer key encryption · retention policies · audit · provenance Native to the data layer, not a layer bolted on. Every read and write is policy-gated for access and provenance; retention runs across the data lifecycle. ### _Konig_ entities · facts & edges · decision traces · episodes Temporal context graph with provenance — sub-200ms retrieval at scale. How they compare ## HydraDB vs. Zep, side by side HydraDB Zep Approach Temporal context graph + vector data layer Bi-temporal temporal context graph (Graphiti) Retrieval Graph + vector (“PageRank for memory”) Unified vector + BM25 + graph traversal + pattern match Temporal model “Git-style” temporal versioning Bi-temporal edges; automatic fact invalidation; point-in-time queries Open source Proprietary (self-host license on higher tier) Graphiti (graph library) is open source Published benchmarks LongMemEval-S 90.79% (Gemini-3); also BEAM, FinanceBench 94.7% LoCoMo, 90.2% LongMemEval (+ peer-reviewed paper) Latency (claimed) <200ms Sub-200ms p95 (published with token figures) Enterprise governance SOC 2, ISO 27001; multi-tenancy; observability ABAC, retention + legal hold, audit; SOC 2 Type II, HIPAA Deployment Managed; self-host license; BYOC (Enterprise) Managed, BYOK, or BYOC (AWS/GCP/Azure) Track record Public beta (2026); $6.5M funded; ~1M retrievals/mo (self-reported) Established; Fortune 500 deployments; S&P coverage A note on the benchmark comparison ## Read the benchmark table carefully HydraDB's published table (its _cortex.pdf_research note) reports 90.79% on LongMemEval-S and lists Zep at 71.2%. Several caveats — most of them visible in HydraDB's own paper — matter before treating that as a head-to-head: - **The backbones differ, by their own labeling.** HydraDB's 90.79% is on **Gemini 3.0 Pro**; the table marks the Zep 71.2% figure as **GPT-4o**. A Gemini-3 result and a GPT-4o result aren't comparable. And 71.2% is Zep's _original 2025-paper_number — not Zep's current published results (90.2% LongMemEval, 94.7% LoCoMo). - **No disclosed context-token budget.** The paper describes retrieval as a “dynamically determined” budget bounded by “context window constraints,” but publishes **no token figure**— so there's nothing to compare against the ~4,408-token context Zep reports. - **No reproducible benchmark code.** HydraDB is proprietary and **publishes no runnable benchmark code**— only a research PDF and a client SDK. Its numbers can't be independently reproduced or code-inspected. Compare current, matched-backbone numbers, and weigh accuracy alongside latency and context-token cost, before drawing conclusions. When to choose ## Pick the tool that fits the problem When HydraDB might fit You want to experiment with a context-graph memory database that combines Git-style temporal versioning with vector recall, and you're comfortable evaluating a newer product as it matures. - Combining Git-style temporal versioning with vector recall - Comfortable evaluating a newer product as it matures - The architectural direction overlaps with where the category is heading When teams choose Zep instead You need a memory layer you can put into production now and operate at scale. - Bi-temporal reasoning with provenance - Governance in the data layer (ABAC, retention, audit) - SOC 2 Type II, HIPAA, and deployment control (managed, BYOK, BYOC) - Published benchmark results and a Fortune 500 track record - S&P Global Market Intelligence coverage of Zep For mission-critical agents, that operational maturity is usually the deciding factor — read the S&P Global Market Intelligence [coverage](/research/sp-global-market-intelligence-zep-coverage/) and the [temporal knowledge graph](/ai-agents/temporal-knowledge-graph/) primer. Get started ## Put governed agent memory into _production_ FAQ ## Frequently asked questions ### How is HydraDB different from Zep? Both use a temporal context graph for agent memory. The practical differences are maturity and operations: Zep is an established managed platform with published benchmarks, enterprise governance, proven scale, and analyst validation; HydraDB is a newer entrant. Re-check current capabilities before deciding. ### Do both handle change over time? Both describe temporal handling. Zep's model is bi-temporal with automatic fact invalidation and point-in-time queries; confirm HydraDB's current temporal semantics against its docs. ### Which is better for enterprise / production? Zep is built and proven for governed memory at enterprise scale (SOC 2 Type II, HIPAA, BYOK/BYOC, millions of graphs). For mission-critical use, that track record lowers risk. ### Is Zep open source? Graphiti — Zep's temporal-context-graph library — is open source. Zep's managed platform, including Konig, its proprietary graph database service, is commercial. --- ## Privacy policy **Source:** https://www.getzep.com/legal/privacy/ [**We're hiring!** Come build with us →](/careers/) **Version 1.0** **Last revised on: January 27, 2024** Zep Software, Inc. (the "Company") is committed to maintaining robust privacy protections for its users.  Our Privacy Policy ("Privacy Policy") is designed to help you understand how we collect, use and safeguard the information you provide to us and to assist you in making informed decisions when using our Service. For purposes of this Agreement, "Site" refers to the Company's website properties, which can be accessed at the getzep.com Internet domain. "Service" refers to the Company's services accessed via the Site, in which users can view Company marketing material, register for the Company's services, access support and help resources, and other services and resources that may be made available from time to time. The terms "we," "us," and "our" refer to the Company. "You" refers to you, as a user of our Site or our Service. By accessing our Site or our Service, you accept our Privacy Policy and [Terms of Use](website-terms-of-use), and you consent to our collection, storage, use and disclosure of your Personal Information as described in this Privacy Policy. 1. INFORMATION WE COLLECT We collect "Non-Personal Information" and "Personal Information." **Non-Personal Information** includes information that cannot be used to personally identify you, such as anonymous usage data, general demographic information we may collect, referring/exit pages and URLs, platform types, preferences you submit and preferences that are generated based on the data you submit and number of clicks. **Personal Information** includes your email and name which you submit to us through the registration process at the Site. 1.1. _Information collected via Technology_ To activate the Service you do not need to submit any Personal Information other than your email address and name. To use the Service thereafter, you do not need to submit further Personal Information. However, in an effort to improve the quality of the Service, we track information provided to us by your browser or by our software application when you view or use the Service, such as the website you came from (known as the "referring URL"), the type of browser you use, the device from which you connected to the Service, the time and date of access, and other information that does not personally identify you. We track this information using cookies, or small text files which include an anonymous unique identifier. Cookies are sent to a user's browser from our servers and are stored on the user's computer hard drive. Sending a cookie to a user's browser enables us to collect Non-Personal information about that user and keep a record of the user's preferences when utilizing our services, both on an individual and aggregate basis. For example, the Company may use cookies to collect the following information: - how often you use our websites and services - which content and features you use The Company may use both persistent and session cookies; persistent cookies remain on your computer after you close your session and until you delete them, while session cookies expire when you close your browser. 1.2. _Information you provide us by registering for an account_ In addition to the information provided automatically by your browser when you visit the Site, to become a subscriber to the Service you will need to create a personal profile. You can create a profile by registering with the Service and entering your email address, and creating a user name and a password. By registering, you are authorizing us to collect, store and use your email address in accordance with this Privacy Policy. 1.3. _Children's Privacy_ The Site and the Service are not directed to anyone under the age of 13. The Site does not knowingly collect or solicit information from anyone under the age of 13, or allow anyone under the age of 13 to sign up for the Service. In the event that we learn that we have gathered personal information from anyone under the age of 13 without the consent of a parent or guardian, we will delete that information as soon as possible. If you believe we have collected such information, please contact us at info@getzep.com. 2. HOW WE USE AND SHARE INFORMATION _Personal Information:_ Except as otherwise stated in this Privacy Policy, we do not sell, trade, rent or otherwise share for marketing purposes your Personal Information with third parties without your consent. We do share Personal Information with vendors who are performing services for the Company, such as the servers for our email communications who are provided access to user's email address for purposes of sending emails from us. Those vendors use your Personal Information only at our direction and in accordance with our Privacy Policy. In general, the Personal Information you provide to us is used to help us communicate with you. For example, we use Personal Information to contact users in response to questions, solicit feedback from users, provide technical support, and inform users about promotional offers. We may share Personal Information with outside parties if we have a good-faith belief that access, use, preservation or disclosure of the information is reasonably necessary to meet any applicable legal process or enforceable governmental request; to enforce applicable Terms of Service, including investigation of potential violations; address fraud, security or technical concerns; or to protect against harm to the rights, property, or safety of our users or the public as required or permitted by law. _Non-Personal Information_ In general, we use Non-Personal Information to help us improve the Service and customize the user experience. We also aggregate Non-Personal Information in order to track trends and analyze use patterns on the Site. This Privacy Policy does not limit in any way our use or disclosure of Non-Personal Information and we reserve the right to use and disclose such Non-Personal Information to our partners, advertisers and other third parties at our discretion. In the event we undergo a business transaction such as a merger, acquisition by another company, or sale of all or a portion of our assets, your Personal Information may be among the assets transferred. You acknowledge and consent that such transfers may occur and are permitted by this Privacy Policy, and that any acquirer of our assets may continue to process your Personal Information as set forth in this Privacy Policy. If our information practices change at any time in the future, we will post the policy changes to the Site so that you may opt out of the new information practices. We suggest that you check the Site periodically if you are concerned about how your information is used. 3. HOW WE PROTECT INFORMATION We implement security measures designed to protect your information from unauthorized access. Your account is protected by your account password and we urge you to take steps to keep your personal information safe by not disclosing your password and by logging out of your account after each use. We further protect your information from potential security breaches by implementing certain technological security measures including encryption, firewalls and secure socket layer technology. However, these measures do not guarantee that your information will not be accessed, disclosed, altered or destroyed by breach of such firewalls and secure server software. By using our Service, you acknowledge that you understand and agree to assume these risks. 4. YOUR RIGHTS REGARDING THE USE OF YOUR PERSONAL INFORMATION You have the right at any time to prevent us from contacting you for marketing purposes.  When we send a promotional communication to a user, the user can opt out of further promotional communications by following the unsubscribe instructions provided in each promotional e-mail. Please note that notwithstanding the promotional preferences you indicate by either unsubscribing, we may continue to send you administrative emails including, for example, periodic updates to our Privacy Policy. 5. LINKS TO OTHER WEBSITES As part of the Service, we may provide links to or compatibility with other websites or applications. However, we are not responsible for the privacy practices employed by those websites or the information or content they contain. This Privacy Policy applies solely to information collected by us through the Site and the Service. Therefore, this Privacy Policy does not apply to your use of a third party website accessed by selecting a link on our Site or via our Service. To the extent that you access or use the Service through or on another website or application, then the privacy policy of that other website or application will apply to your access or use of that site or application. We encourage our users to read the privacy statements of other websites before proceeding to use them. 6. CHANGES TO OUR PRIVACY POLICY The Company reserves the right to change this policy and our Terms of Service at any time.  We will notify you of significant changes to our Privacy Policy by sending a notice to the primary email address specified in your account or by placing a prominent notice on our site. Significant changes will go into effect 30 days following such notification. Non-material changes or clarifications will take effect immediately. You should periodically check the Site and this privacy page for updates. 7. CONTACT US If you have any questions regarding this Privacy Policy or the practices of this Site, please contact us by sending an email to info@getzep.com. --- ## Sub-processors **Source:** https://www.getzep.com/legal/subprocessors/ [**We're hiring!** Come build with us →](/careers/) Legal Third parties Zep engages to operate the Context Lake, agent memory, and related platform services. Last updated: August 17, 2026 Sub-processor Country of processing Purpose of processing Amazon Web Services, Inc. United States Cloud infrastructure services used to host and operate Zep's platform, including storage of customer data and execution of application workloads. Microsoft Corporation (Azure) United States Cloud compute and storage services supporting model inference, data processing, and operation of Zep-managed workloads. OpenAI, L.L.C. United States Large-language-model processing used to transform customer-submitted data, including extraction, summarization, and generation tasks. Hatchet Technologies, Inc. United States Managed job-orchestration services used for scheduling and executing background processing tasks. Kapa.ai Inc. United States AI-powered documentation and support services used to provide customer assistance and knowledge-base interactions. Kinde Australia Pty Ltd United States Authentication and user-identity management services for Zep's customer-facing applications. PostHog Inc. United States Product analytics services used to measure feature usage and application performance. Google LLC (Gemini) United States Large-language-model processing used for summarization, extraction, and related AI transformations of customer-submitted data. Datadog, Inc. United States Observability services used for infrastructure monitoring, log processing, metric collection, and application-level diagnostics. Stripe, Inc. United States Payment processing and billing services used to collect and manage customer subscription payments. Schematic, Inc. United States Plan, entitlement, and usage-metering services used to manage customer subscriptions and feature access. Svix, Inc. United States Webhook delivery infrastructure used to send event notifications to customer endpoints. Loops United States Lifecycle and marketing email services used for customer onboarding and product communications. Resend, Inc. United States Transactional email services used for feedback collection and operational notifications. Cloudflare, Inc. United States Content delivery, web application firewall, and edge services used to protect and accelerate Zep's platform. Reo.dev United States Product and go-to-market analytics services used to measure website and product engagement. Microsoft Corporation (Clarity) United States Session analytics services used to understand website visitor behavior and improve user experience. LinkedIn Corporation United States Advertising and conversion-tracking services used to measure marketing campaign performance. Thinking Machines Lab Inc. United States Large-language-model processing used for extraction, summarization, and related AI transformations of customer-submitted data. Deep Infra, Inc. United States Model inference services used to run large-language-model processing of customer-submitted data, including extraction, summarization, and generation tasks. --- ## Terms of service **Source:** https://www.getzep.com/legal/terms/ [**We're hiring!** Come build with us →](/careers/) **Last updated: August 17, 2026** Zep Software, Inc., a Delaware corporation ("Zep"), operates a cloud-based long-term memory and context-engineering software platform for AI agents and assistants (the "Platform"). These terms govern the access to and use of the Zep Service (as defined below) by the entity or individual entering into this agreement ("Customer"). This Agreement takes effect on the date Customer first accepts these terms, whether by executing an Order Form, clicking "I accept" or "I agree" (or similar) at account creation, or beginning to use the Zep Service (the "Effective Date"). If Customer is accepting on behalf of a company or other legal entity, Customer represents and warrants that it has authority to bind that entity to this Agreement. This Agreement consists of these terms and conditions (the "Terms of Service"), any Order Form(s) referencing these Terms of Service, the subscription plan and usage terms displayed on the Platform, and any addenda executed by authorized representatives of the parties (collectively, the "Agreement"). Customer's access to and use of the Zep Service is governed solely by the Agreement. ## 1\. Definitions Capitalized terms have the meaning set forth below or as defined within these Terms of Service. **1.1 "API Key"** means the unique application programming interface key(s) provisioned by Zep to Customer for purposes of accessing and using the Zep Service programmatically. **1.2 "AI Tools"** means generative artificial intelligence and machine learning services or applications that are integrated into the Zep Service, including without limitation, third-party large language models. **1.3 "Applicable Privacy Laws"** means the data protection, data security and privacy laws and regulations of any jurisdiction applicable to the Zep Service under this Agreement. **1.4 "Confidential Information"** means all information regarding a party's business, including, without limitation, technical, marketing, financial, employee, planning, and other confidential or proprietary information, that (a) is clearly identified as confidential or proprietary at the time of disclosure, or (b) the receiving party knew or should have known, given the nature of the information and the circumstances of its disclosure, was considered confidential or proprietary. **1.5 "Customer Data"** means Inputs, Outputs, and any other content or information uploaded or transmitted to the Zep Service by Customer or Users, including from Third-Party Services. Customer Data does not include Performance Data or Aggregate Data. **1.6 "Documentation"** means all specifications, user manuals, and other technical materials relating to the Zep Service that are provided or made available to Customer, and as may be modified by Zep from time to time. **1.7 "Fees"** means the fees for the Zep Service as set forth on an applicable Order Form or, for self-serve subscriptions, as published on the Platform at the time of Customer's subscription. **1.8 "Free Plan"** means a subscription tier that enables access to and use of the Zep Service on a no-charge basis, as described on the Platform. **1.9 "Order Form"** means an order form executed by the parties and referring to this Agreement which specifies the Zep Service and applicable Fees. **1.10 "Paid Plan"** means any subscription to the Zep Service for which Customer pays Fees, including the Flex, Flex Plus, and Enterprise plans, as described on the Platform. **1.11 "Personal Data"** means Customer Data that constitutes "personal data," "personal information," or "personally identifiable information" defined in Applicable Privacy Laws or information of a similar character regulated thereby, except that Personal Data does not include such information pertaining to Customer personnel who are business contacts of Customer, or such information received by Zep directly or from other sources (such as its other customers) independent of Zep's relationship with Customer. **1.12 "Prohibited Data"** means (a) payment card data subject to the Payment Card Industry Data Security Standard (PCI-DSS), (b) Protected Health Information (as defined by HIPAA), except where Customer and Zep have executed a Business Associate Agreement, and (c) any other categories of data designated as prohibited in the Documentation or on the Platform. **1.13 "Zep Service"** means Zep's proprietary service, including the application programming interfaces, software development kits, and the web-based dashboard made available by Zep to Customer, as further described in the Documentation or an applicable Order Form. **1.14 "Zep Technology"** means the Zep Service, Performance Data, the Aggregate Data, the Documentation, and all applicable software, data, or technical information used by Zep or provided to Customer in connection with the foregoing. **1.15 "Third-Party Service"** means any third-party service or application connected to, or integrated with, the Zep Service by or on behalf of Customer. **1.16 "Users"** means employees, independent contractors, and end users of applications of Customer that are authorized by Customer to access the Zep Service pursuant to Customer's rights under this Agreement, including through the use of API Key(s). ## 2\. Zep Service; Access; Restrictions **2.1 Subscription to the Zep Service.** Subject to the terms and conditions of this Agreement, Zep hereby grants to Customer a revocable, non-sub-licensable, non-transferable (except as provided in Section 15.3), non-exclusive right to access and use the Zep Service and accompanying Documentation solely for Customer's internal business purposes. **2.2 Access.** Customer will access and use the Zep Service through API Key(s) provisioned by Zep and, where applicable, through the web-based dashboard using unique account credentials ("Account"). API Key(s) are confidential and may not be shared with any unauthorized third party. Customer may provision multiple API Key(s) within its account and may use such API Key(s) in Customer's own applications to serve Users. Customer is responsible for maintaining the confidentiality of all API Key(s) and account credentials and is solely responsible for all activities that occur thereunder. Customer is responsible for ensuring that its Users comply with the terms of this Agreement and shall be liable for any acts or omissions of its Users that would constitute a breach of this Agreement. Customer will promptly notify Zep of any actual or suspected unauthorized use or access to its account or API Key(s). **2.3 Restrictions.** Customer will not, and will not permit any User or other party to: (a) allow any third party to access the Zep Technology except as expressly allowed herein; (b) sublicense, lease, sell, resell, rent, loan, distribute, transfer or otherwise allow the use of the Zep Technology for the benefit of any unauthorized third party; (c) reverse engineer, decompile, disassemble, or otherwise derive or determine or attempt to derive or determine the source code (or the underlying ideas, algorithms, structure or organization) of the Zep Technology, except as permitted by law; (d) use any automated software, devices or other processes to "scrape," extract, or download data from the Zep Technology (other than Customer Data) without the prior written consent of Zep; (e) interfere in any manner with the operation of the Zep Technology or the hardware and network used to operate the same, or attempt to probe, scan or test vulnerability of the Zep Technology without the prior written consent of Zep; (f) attempt to access the Zep Technology through any unapproved interface; (g) attempt to circumvent any usage restrictions of the Zep Technology; (h) modify, copy or make derivative works based on any part of the Zep Technology; (i) access or use the Zep Technology to build a similar or competitive product or service, or to develop a product or service that is substantially similar to or competes with the Zep Service, or otherwise engage in competitive analysis or benchmarking for public or third-party disclosure without Zep's prior written consent; provided that Customer may conduct such testing and benchmarking solely for its own internal evaluation purposes; (j) remove, alter, or obscure any proprietary notices (including copyright and trademark notices) of Zep or its licensors on the Zep Technology or any copies thereof; (k) upload or transmit any Prohibited Data to the Zep Service except as expressly authorized in writing by Zep; or (l) otherwise use the Zep Technology in any manner that exceeds the scope of use permitted under Section 2.1 or in a manner inconsistent with applicable law, the Documentation, the Order Form or this Agreement. **2.4 Suspension.** Zep reserves the right to suspend Customer's or any User's access to the Zep Service for any failure, or suspected failure, to comply with the restrictions set forth in Section 2.3. Zep may also suspend Customer's or any User's access to all or any part of the Zep Service, without notice and without incurring any resulting obligation or liability, if Zep believes, in its good faith and reasonable discretion, that Customer's or any User's use of the Zep Service poses a risk to the security or integrity of Zep's systems, interferes with Zep's ability to reliably provide the Zep Service to other customers, or may subject Zep to liability. Zep will use reasonable efforts to notify Customer or the applicable User(s) prior to suspension and will restore access to Customer or the applicable User(s) as soon as such risks no longer apply. **2.5 Customer Data.** Customer will have the sole responsibility for the accuracy, quality, integrity, legality, reliability, and appropriateness of all Customer Data. Customer Data will not: (a) be deceptive, defamatory, obscene, pornographic or unlawful; (b) include any Prohibited Data, except as expressly authorized in writing by Zep; (c) knowingly contain any viruses, worms or other malicious computer programming codes intended to damage the Zep Service; or (d) violate the intellectual property, privacy, or other rights of any third party or violate any Applicable Privacy Laws. **2.6 Third-Party Services.** Customer may elect to link certain Third-Party Services to the Zep Service. Customer is responsible for enabling the integration of each Third-Party Service, and by doing so, Customer acknowledges that: (a) Zep may access any Customer Data provided via a Third-Party Service so that it may be used in accordance with the terms of this Agreement, and (b) it is instructing Zep to share Customer Data (including Personal Data where directed) with the providers of such Third-Party Services. Third-Party Services are not under the control of Zep and Zep is not responsible for any Third-Party Services. Customer's use of the Third-Party Services is governed by the Customer's agreement with providers of the Third-Party Services. Customer acknowledges and agrees that, for the purposes of Applicable Privacy Laws, each of Zep and providers of any Third-Party Service are not processors or subprocessors of Personal Data with respect to each other. **2.7 Use of AI Tools.** The Zep Service may include AI Tools. Customer may submit queries or other Customer Data to the AI Tools ("Inputs") and receive back outputs generated by the AI Tools in response to Customer's Inputs ("Outputs"). Inputs and Outputs are both Customer Data. Inputs will be shared with Third-Party Services that provide the AI Tools in order to generate Outputs. Zep's rights to use Inputs, Outputs, and related metadata for training or improving the AI Tools are subject to the terms set forth in Section 5.2. Customer acknowledges and agrees that Zep does not represent or warrant that Outputs will (a) be free from third-party content or (b) not infringe third-party intellectual property rights. Customer acknowledges that the services leverage AI Tools and that Zep is not liable, and Customer agrees not to seek to hold Zep liable, for any third-party AI Tools. Customer is solely responsible for ensuring that its and its Users' use of the Zep Service and Outputs comply with all applicable laws. Customer will be solely responsible for Customer's and its Users' use of the Zep Service and any Outputs resulting therefrom. Customer should evaluate the fitness of any Output as appropriate for Customer's specific use case. **2.8 Service Modifications.** Zep reserves the right to modify, update, or discontinue any feature or functionality of the Zep Service at any time. If Zep makes a change that materially reduces the core functionality of the Zep Service as described in the applicable Order Form during the then-current Subscription Term, Zep will provide Customer with at least thirty (30) days' prior written notice. If such change materially and adversely affects Customer's use of the Zep Service under a Paid Plan, Customer may terminate the affected Order Form within thirty (30) days of receiving such notice, and Zep will refund any prepaid Fees attributable to the remainder of the then-current Subscription Term. ## 3\. Support The terms of this Section 3 apply only to Customers accessing or using the Platform pursuant to a Paid Plan. Subject to the terms and conditions of this Agreement, Zep will exercise commercially reasonable efforts to: (a) provide support to Customer for the use of Zep Service; and (b) keep the Zep Service operational and available to Customer, in each case in accordance with industry standards and its standard policies and procedures. ## 4\. Fees and Payment **4.1 Fees.** Customer will pay Zep the applicable Fees. For self-serve subscriptions, Fees are prepaid via credit card or other payment method accepted by Zep and are charged at the beginning of each billing period at the rates then in effect. Zep may update the Fees applicable to self-serve subscriptions at any time by publishing revised rates on the Platform; provided that Zep will provide Customer with at least thirty (30) days' prior notice (which may be provided via email or through the Platform) before any such revised Fees take effect. Any revised Fees will apply beginning with the first billing period commencing after the effective date of the change; Fees for the then-current billing period will not be affected. Usage-based overages beyond applicable plan limits will be billed in accordance with the rates published on the Platform or set forth in the applicable Order Form. Notwithstanding the foregoing, for Enterprise tier subscriptions: (a) Customer will pay the Fees set forth on the applicable Order Form within thirty (30) days of receipt of an invoice; (b) all Fees are non-refundable (except as expressly set out in this Agreement or an Order Form) and are not eligible for set off; (c) Customer will maintain complete, accurate and up-to-date Customer billing and contact information; and (d) unless otherwise stated on an Order Form, at the end of the Initial Term or any subsequent Renewal Term, Zep reserves the right to increase the Fees payable for the forthcoming Renewal Term upon written notice to Customer at least sixty (60) days prior to the commencement of the Renewal Term and such revised Fees will take effect immediately upon the commencement of the Renewal Term. **4.2 Taxes.** All Fees owed by Customer in connection with this Agreement are exclusive of, and Customer will pay, all sales, use, excise and other taxes and applicable export and import fees, customs duties and similar charges that may be levied upon Customer in connection with this Agreement, except for employment taxes and taxes based on Zep's income. **4.3 Late Payment.** Payments by Customer that are past due will be subject to interest at the rate of one and one-half percent (1.5%) per month (or, if less, the maximum allowed by applicable law) of that overdue balance. Zep reserves the right (in addition to any other rights or remedies Zep may have) to suspend Customer's access to the Zep Service if any Fees are more than fifteen (15) days overdue until such amounts are paid in full. Notwithstanding the foregoing, if any Fees remain unpaid for more than fifteen (15) days after written notice of such nonpayment from Zep, Zep may terminate this Agreement or the applicable Order Form upon written notice to Customer. ## 5\. Proprietary Rights **5.1 Zep Technology.** Customer acknowledges that Zep retains all right, title and interest in and to the Zep Technology, including any enhancements, improvements, or derivatives thereto, and that the Zep Technology is protected by intellectual property rights owned by or licensed to Zep. Other than as expressly set forth in this Agreement, no license or other rights in the Zep Technology are granted to the Customer. **5.2 Customer Data.** Customer retains all right, title and interest in and to Customer Data. Customer hereby grants to Zep a non-exclusive, worldwide, perpetual, irrevocable, royalty-free and fully paid-up license to access, use, reproduce, modify, and create derivative works from Customer Data for any lawful purpose, including to provide the Zep Service, to train and improve machine learning and artificial intelligence models, and to develop and improve Zep's products and services. **5.3 Aggregate Data.** Notwithstanding Section 5.2, Zep may create aggregated and de-identified data derived from Customer Data and Performance Data ("Aggregate Data"). Zep shall own all right, title, and interest in and to the Aggregate Data, and may use such Aggregate Data for any purpose, including to improve the Zep Service and to develop and improve Zep's products and services. Aggregate Data will not identify Customer or any individual. **5.4 Performance Data.** Zep may monitor Customer's use of the Zep Service and may collect and compile general performance and usage data about the Zep Service, including Customer's use of the Zep Service (such as technical logs) ("Performance Data"). Performance Data does not include any Customer Data. As between Zep and Customer, all right, title, and interest in the Performance Data, and all intellectual property rights therein, belong to and are retained solely by Zep. Zep may use Performance Data for any purpose, provided that any disclosure or use of Performance Data outside of Zep's internal operations will be in aggregated and de-identified form and will not identify Customer or Customer's Confidential Information. **5.5 Feedback.** Customer or its Users may give feedback to Zep on the use, operation, and functionality of the Zep Service, including information about operating results, known or suspected bugs, errors, or compatibility problems, suggested modifications, and user-desired features, functionality, or workflows (collectively, "Feedback"). Customer hereby grants Zep a perpetual, irrevocable, worldwide, royalty-free and fully paid-up license to use, reproduce, modify, and create derivative works of the Feedback in connection with its business, products and services without restriction or consideration to Customer. Zep will not identify Customer as the source of any such feedback. Zep acknowledges that all Feedback is provided to Zep on an "as is" basis and that Customer is not responsible for Zep's use of any Feedback, including any results therefrom. ## 6\. Data Security The terms of this Section 6 apply only to Customers accessing or using the Platform pursuant to a Paid Plan. During the Term, Zep will implement and maintain commercially reasonable administrative, technical and physical measures designed to safeguard against unauthorized access to or use or disclosure of any Customer Data or Personal Data. Customer and its Users will be responsible for all changes to and/or deletions of Customer Data and the security of all passwords and other usernames and passwords required to access the Zep Service. In the event Zep becomes aware of any loss or unauthorized access, disclosure or use of any Personal Data ("Security Incident"), Zep will (a) promptly notify Customer in writing of such Security Incident, and (b) take commercially reasonable steps designed to (i) identify the cause of such Security Incident, (ii) minimize the harm associated therewith and (iii) prevent reoccurrence thereof. Any notification of any Security Incident will describe, to the extent known, details of the Security Incident, including steps taken by Zep, or that Zep recommends that Customer take, to mitigate the potential risks. Zep's notification of or response to a Security Incident will not be construed as Zep's acknowledgment of any fault or liability with respect to the Security Incident. ## 7\. Privacy Zep will process Personal Data only in accordance with Customer's instructions to Zep contained in the Agreement. This Agreement is a complete expression of such instructions, and Customer's additional instructions will be binding on Zep only pursuant to an amendment to this Agreement signed by both parties. By entering into this Agreement, Customer instructs Zep to process Personal Data to provide the Zep Service and to perform its other obligations and exercise its rights under the Agreement. Customer will ensure (and is solely responsible for ensuring) that it has given such notices to and obtained such consents and permissions from all relevant third parties, and has reserved all rights, in each case, as may be required under applicable law or otherwise for Zep to process Personal Data as contemplated by the Agreement. If Customer and Zep execute a Data Processing Addendum ("DPA"), upon mutual execution, the DPA will be incorporated into and form part of this Agreement. In the event of a conflict between the terms of this Agreement and the DPA, the DPA will govern and control with respect to the processing of Personal Data. ## 8\. HIPAA; Business Associate In providing the Zep Service hereunder, Zep may be considered a "business associate" of Customer as defined under the Health Insurance Portability and Accountability Act of 1996, as amended, and the implementing rules and regulations thereunder related to privacy, security and breach notification ("HIPAA"). Customer acknowledges that the transmission of Protected Health Information (as defined by HIPAA) to the Zep Service constitutes Prohibited Data unless and until Customer and Zep have executed a Business Associate Agreement ("BAA"). If Customer intends to use the Zep Service in connection with Protected Health Information, Customer must first execute Zep's standard form BAA. Upon mutual execution, the BAA will be incorporated into and form part of this Agreement. In the event of a conflict between the terms of this Agreement and the BAA, the BAA will govern and control with respect to the use and protection of Protected Health Information. ## 9\. Confidential Information **9.1 Restrictions.** As a recipient of Confidential Information, each party will (a) use the Confidential Information of the disclosing party only as set forth in this Agreement, (b) not disclose to any third party any Confidential Information of the disclosing party, except as expressly permitted under this Agreement, (c) limit access to the Confidential Information of the disclosing party to its employees and contractors who have a need to know such information to use or provide the Zep Service, and ensure that such employees or contractors are bound by confidentiality obligations at least as protective as those contained herein, and (d) protect the Confidential Information of the disclosing party from unauthorized use, access, or disclosure in a reasonable manner. **9.2 Exclusions.** The restrictions on use and disclosure of Confidential Information set forth above will not apply to any Confidential Information that (a) is or becomes generally known and available to the public through no act or omission of the receiving party, (b) was in the receiving party's lawful possession without confidentiality restrictions prior to disclosure by the disclosing party, (c) is received without confidentiality restrictions from a third party with the right to make such a disclosure, or (d) is independently developed by the receiving party. The receiving party may disclose Confidential Information to the extent that such disclosure is required by law or by the order of a court or similar judicial or administrative body, provided that the receiving party will, if permitted by law, provide advance notice of the disclosure to the disclosing party and cooperate so that the disclosing party has the opportunity to obtain appropriate confidential treatment for such Confidential Information. ## 10\. Term and Termination **10.1 Term.** The term of this Agreement will commence on the Effective Date and continue until all Order Forms or subscriptions have expired or been terminated, unless terminated earlier in accordance with the terms of this Agreement (the "Term"). For self-serve subscriptions, the subscription term will be monthly or annual as selected by Customer at the time of subscription and will automatically renew for successive periods of the same duration, unless Customer cancels prior to the end of the then-current billing period. For enterprise subscriptions, unless otherwise set forth in an Order Form, each Order Form will have an initial term of one (1) year (the "Initial Term") and will automatically renew for successive one (1) year terms (each a "Renewal Term" and together with the Initial Term, the "Subscription Term"), unless either party provides written notice of its intent to terminate the Order Form at least thirty (30) days prior to the end of the then-current term. Customers on a Free Plan may use the Zep Service on an at-will basis, subject to the terms of this Agreement, and either party may terminate such access at any time. **10.2 Termination.** Either party may terminate this Agreement upon written notice if: (a) the other party materially breaches the Agreement and does not cure such breach (if curable) within thirty (30) days after written notice of such breach, or (b) the other party: (i) becomes insolvent, (ii) files a petition in bankruptcy that is not dismissed within sixty (60) days of commencement, or (iii) makes an assignment for the benefit of its creditors. **10.3 Effect of Termination.** Upon the expiration or termination of this Agreement for any reason, the rights and licenses granted to Customer hereunder will immediately terminate and Customer will cease use of the Zep Service and Documentation. Termination of this Agreement will not relieve Customer of its obligation to pay all Fees that accrued prior to such termination. Each party will return to the other or destroy all property (including any Confidential Information) of the other party. Notwithstanding the foregoing, (a) each party may retain the Confidential Information of the other in accordance with its standard backup procedures, subject to the requirements in Section 9 (Confidential Information) and Section 6 (Data Security), and (b) Zep's rights in Aggregate Data and Performance Data as set forth in Sections 5.3 and 5.4 will survive termination. Sections 1, 2.3, 2.6, 2.7, 4, 5 (excluding any term-limited license grants), 6, 7, 8, 9, 10.3, and 11-15 will survive the termination of this Agreement. ## 11\. Limited Warranties Customer represents and warrants that it has all rights necessary to upload and use Customer Data (including any Customer Data uploaded or transmitted by its Users) with the Zep Service and to grant Zep all licenses to Customer Data in this Agreement without violating any third-party intellectual property, privacy or other rights, including Applicable Privacy Laws. During the Term, Zep warrants that the Zep Service, when used in accordance with the Documentation and the terms of this Agreement, will operate as described in the Documentation in all material respects. If Customer notifies Zep of any breach of the foregoing warranty, Zep will, as Customer's sole and exclusive remedy, use commercially reasonable efforts to repair and fix the non-conforming service. ## 12\. Disclaimer Except as expressly provided herein, and to the maximum extent permitted by applicable law: (a) the Zep Technology is provided "AS-IS" and "AS AVAILABLE" and (b) Zep and its suppliers make no other warranties, express or implied, by operation of law or otherwise, and hereby expressly disclaim any and all other warranties including, without limitation, any implied warranties of merchantability, fitness for a particular purpose, title, or non-infringement. Zep does not warrant or represent that the Zep Technology will be free from bugs or uninterrupted or error-free, or make any other representations regarding the use, or the results of the use, of the Zep Technology in terms of correctness, accuracy, reliability, or otherwise. Without limiting the foregoing, Zep makes no warranty that any Outputs or other results generated by the AI Tools will be accurate, complete, reliable, non-infringing, or fit for any particular purpose. Customer acknowledges that AI-generated outputs may contain errors, omissions, or inaccuracies and that Customer is solely responsible for evaluating and verifying all outputs before use. Customer acknowledges and agrees that Zep is not liable, and Customer agrees it will not seek to hold Zep liable, for the conduct of third parties, including any Third-Party Service, and that the risk of injury from any third party rests entirely with Customer. ## 13\. Indemnity **13.1 By Zep.** The terms of this Section 13.1 apply only to Customers accessing or using a Paid Plan. If any action is instituted by a third party against Customer based upon a claim that the Zep Technology, as delivered and when used in accordance with this Agreement, infringes any third party's intellectual property rights, Zep will defend such action at its own expense on behalf of Customer and will pay all damages attributable to such claim that are finally awarded against Customer or paid in settlement. The foregoing indemnification obligation does not apply to alleged infringement or misappropriation arising from Outputs. If the Zep Technology is enjoined or, in Zep determination is likely to be enjoined, Zep will, at its option and expense (a) procure for Customer the right to continue using the Zep Technology, (b) replace or modify the Zep Technology so that it is no longer infringing but continues to provide comparable functionality, or (c) terminate this Agreement and Customer's access to the Zep Technology and refund any amounts previously paid for the Zep Technology attributable to the remainder of the then-current Subscription Term. Zep will have no obligation under this Section 13.1 or otherwise with respect to any infringement claim based upon: (i) any use of the Zep Technology not in accordance with this Agreement or as specified in the Documentation; (ii) any use of the Zep Technology in combination with other products, equipment, software or data not supplied by Zep, including Third-Party Services; or (iii) any modification of the Zep Technology by any person other than Zep or its authorized agents. This Section 13.1 sets forth the entire obligation of Zep and the exclusive remedy of Customer against Zep for any claim that the Zep Technology infringes a third party's intellectual property rights. **13.2 By Customer.** If any action is instituted by a third party against Zep relating to (a) Customer Data (including any Customer Data uploaded or transmitted by Users), (b) any act or omission of Customer's Users in connection with the Zep Service, or (c) Customer's breach or alleged breach of Section 2.5 or Customer's representations and warranties set forth in Section 11, Customer will defend such action at its own expense on behalf of Zep and will pay all damages attributable to such claim that are finally awarded against Zep or paid in settlement of such claim. **13.3 Procedure.** Any party that is seeking to be indemnified under the provision of this Section 13 (the "Indemnified Party") must (a) promptly notify the other party (the "Indemnifying Party") of any third-party claim, suit, or action for which it is seeking an indemnity hereunder (a "Claim"), (b) give the Indemnifying Party the sole control over the defense of such Claim, and (c) reasonably cooperate with the Indemnifying Party at the Indemnifying Party's expense. The Indemnifying Party will not agree to any settlement that requires the Indemnified Party to admit to fault or to take or refrain from taking any action without the Indemnified Party's prior written consent. ## 14\. Limitation of Liability To the extent permitted by law, in no event will Zep be liable to Customer for special, incidental, consequential or punitive damages or lost profits in any way relating to this Agreement. In no event will Zep's aggregate, cumulative liability to Customer in any way relating to this Agreement exceed the greater of (a) the amount of fees actually received by Zep from Customer pursuant to the applicable Order Form during the twelve (12) months preceding the claim; and (b) $100. The foregoing limitations will not apply to liabilities that cannot be limited by law. The parties would not have entered into this Agreement but for such limitations. ## 15\. General Provisions **15.1 Governing Law.** This Agreement will be governed by, and all disputes arising under or in connection with this Agreement will be resolved in accordance with, the laws of the State of California, United States of America, exclusive of conflict or choice of law rules. **15.2 Dispute Resolution.** All disputes arising out of or in connection with this Agreement, including any question regarding its formation, existence, validity or termination, will be finally settled under the Commercial Arbitration Rules of the American Arbitration Association (the "Arbitration Rules") by one or more arbitrators appointed in accordance with the said Arbitration Rules. The seat, or legal place, of the arbitration will be San Francisco, California, United States of America. The language of the arbitration will be English. Except as otherwise specifically limited in this Agreement, the arbitral tribunal will have the power to grant any remedy or relief that it deems appropriate, whether provisional or final, including but not limited to conservatory relief and injunctive relief. Each party retains the right to apply to any court of competent jurisdiction for interim and/or conservatory measures, including pre-arbitral attachments or preliminary injunctions, and any such request will not be deemed incompatible with, or a waiver of, this agreement to arbitrate. The existence and content of the arbitral proceedings and any rulings or awards will be kept confidential by the parties and members of the arbitral tribunal except (a) to the extent that disclosure may be required of a party to fulfill a legal duty, protect or pursue a legal right, or enforce or challenge an award in bona fide legal proceedings before a state court or other judicial authority, (b) with the consent of all parties, (c) where needed for the preparation or presentation of a claim or defense in this arbitration, (d) where such information is already in the public domain other than as a result of a breach of this clause, or (e) by order of the arbitral tribunal upon application of a party. The arbitration award will be final and binding on the parties, and the parties undertake to carry out any award without delay. The parties will be deemed to have waived their right to any form of recourse insofar as such waiver can validly be made. Judgment on the award may be entered in any court of competent jurisdiction. **15.3 Assignment; Subcontractors.** Neither party may assign this Agreement, including any rights or obligations arising hereunder, without the prior written consent of the other, except that either party may assign this Agreement without the consent of the other party in connection with a merger, acquisition, corporate reorganization, or sale of all or substantially all of its assets. Any attempted assignment or transfer in violation of the foregoing will be null and void. This Agreement will be binding upon each party's respective permitted successors and assigns. Customer agrees that Zep may subcontract certain aspects of the Zep Service to qualified third parties, provided that any such subcontracting arrangement will not relieve Zep of any of its obligations hereunder. **15.4 Order of Precedence.** In the event of a conflict between the Terms of Service, an Order Form, or an exhibit to the Agreement, the following order of precedence will govern: an Order Form (as applicable), the Terms of Service, and then the other exhibits, if any. Notwithstanding the foregoing, an Order Form will take precedence over the Terms of Service if the Order Form expressly states which sections of these Terms of Service are intended to be superseded by the Order Form. **15.5 Notices.** Any notice under this Agreement must be given in writing to the other party by email. Notices to Zep must be sent to notices@getzep.com and notices to Customer will be sent to the email address associated with Customer's Account or, if applicable, the email address specified in an Order Form. Notices will be deemed to have been given on the date sent by email, provided the sender does not receive an automated non-delivery notification. Each party is responsible for keeping its notice email address current and will notify the other party of any change by email to the then-current notice address. To be deemed effective, any email notice of the other party's material breach pursuant to Section 10.2 must reference Section 10.2. **15.6 Force Majeure.** Any delay in the performance of any duties or obligations of either party (except for the obligation to pay Fees owed) will not be considered a breach of this Agreement if such delay is caused by a labor dispute, shortage of materials, war, fire, earthquake, typhoon, flood, natural disasters, governmental action, pandemic/epidemic, cloud-service provider outage, or any other event beyond the control of such party (collectively, a "Force Majeure Event"), provided that such party uses reasonable efforts, under the circumstances, to notify the other party of the circumstances causing the delay and to resume performance as soon as possible. If the Zep Service is unavailable or materially degraded for a continuous period of fifteen (15) days due to a Force Majeure Event, either party will have the right to terminate the Agreement, and Zep will refund any amounts previously paid for the Zep Service attributable to the remainder of the then-current Subscription Term. **15.7 Publicity.** Zep may use Customer's name and logo to identify Customer as a customer, including on Zep's website, social media and in sales and marketing materials, in the same manner in which it uses the names of its other customers. Zep will use Customer's name and logo in accordance with Customer's applicable branding guidelines and Zep may not use Customer's name or logo in any other way without Customer's prior written consent. Customer may opt out of the foregoing right at any time by providing written notice to Zep (including via email), and Zep will remove Customer's name and logo from its marketing materials within a commercially reasonable period following receipt of such notice. **15.8 Export.** Customer agrees not to use, export, re-export, or transfer, directly or indirectly, any U.S. technical data acquired from Zep, or any products utilizing such data, in violation of the United States export laws or regulations. Further, each party agrees to comply with all relevant export laws and regulations of the United States and the country or territory in which the Zep Service is provided ("Export Laws") to assure that neither any deliverable, if any, nor any direct product thereof is (a) exported, directly or indirectly, in violation of Export Laws or (b) intended to be used for any purposes prohibited by the Export Laws, including without limitation nuclear, chemical, or biological weapons proliferation. Customer further represents that (i) Customer is not located in a country that is subject to a U.S. Government embargo, or that has been designated by the U.S. Government as a "terrorist supporting" country and (ii) Customer is not listed on any U.S. Government list of prohibited or restricted parties. Customer acknowledges and agrees that products, services or technology provided by Zep are subject to the export control laws and regulations of the United States, agrees to comply with these laws and regulations, and agrees that it will not, without prior U.S. government authorization, export, re-export, or transfer Zep's products, services or technology, either directly or indirectly, to any country in violation of such laws and regulations. **15.9 U.S. Government Restricted Rights.** If Customer is a government end user, then this provision also applies to Customer. The software contained within the Zep Service and provided in connection with this Agreement has been developed entirely at private expense, as defined in FAR section 2.101, DFARS section 252.227-7014(a)(1) and DFARS section 252.227-7015 (or any equivalent or subsequent agency regulation thereof), and is provided as "commercial items," "commercial computer software" and/or "commercial computer software documentation." Consistent with DFARS section 227.7202 and FAR section 12.212, and to the extent required under U.S. federal law, the minimum restricted rights as set forth in FAR section 52.227-19 (or any equivalent or subsequent agency regulation thereof), any use, modification, reproduction, release, performance, display, disclosure or distribution thereof by or for the U.S. Government will be governed solely by this Agreement and will be prohibited except to the extent expressly permitted by this Agreement. **15.10 Miscellaneous.** This Agreement (as modified by the parties from time to time) constitutes the entire understanding and agreement of the parties and supersedes all prior and contemporaneous understandings. Only a written amendment signed by both parties may modify this Agreement; provided, however, that Zep may modify these Terms of Service for Customers on a Free Plan or a self-serve Paid Plan by posting updated terms on the Platform and providing at least thirty (30) days' prior notice (via email or through the Platform), and Customer's continued use of the Zep Service after the effective date of any such modification will constitute Customer's acceptance of the modified terms. If any provision of this Agreement is held to be invalid or unenforceable, the valid or enforceable portion thereof and the remaining provisions of this Agreement will remain in full force and effect. Any waiver or failure to enforce any provision of this Agreement on one occasion will not be deemed a waiver of any other provision or of such provision on any other occasion. All waivers must be in writing. The headings of Sections of this Agreement are for convenience and are not to be used in interpreting this Agreement. As used in this Agreement, the word "including" means "including but not limited to." There are no third-party beneficiaries of this Agreement. The parties to this Agreement are independent contractors, and no agency, partnership, franchise, joint venture or employee-employer relationship is intended or created by this Agreement. Prior version: [Terms of Service last updated August 7, 2025](/legal/terms/2025-08-07/) --- ## Terms of service (August 7, 2025) **Source:** https://www.getzep.com/legal/terms/2025-08-07/ [**We're hiring!** Come build with us →](/careers/) This is a prior version of Zep's Terms of Service, last updated August 7, 2025. See the [current Terms of Service](/legal/terms/). **Version 1.1** **Last revised on: August 7, 2025** Changelog: v1.1: Addition of _Benchmarking Restrictions_ * * * If you signed a separate Cover Page to access the Product with the same account, and that agreement has not ended, the terms below do not apply to you. Instead, your separate Cover Page applies to your use of the Product. This Agreement is between Zep Software, Inc. and the company or person accessing or using the Product. This Agreement consists of: (1) the Order Form and (2) the Key Terms, both of which are on the Cover Page below, and (3) the Common Paper [Cloud Service Agreement Standard Terms Version 1.1](https://commonpaper.com/standards/cloud-service-agreement/1.1/) ("Standard Terms"). Any modifications to the Standard Terms made in the Cover Page will control over conflicts with the Standard Terms. Capitalized words have the meanings or descriptions given in the Cover Page or the Standard Terms. If you are accessing or using the Product on behalf of your company, you represent that you are authorized to accept this Agreement ßon behalf of your company. By signing up, accessing, or using the Product, Customer indicates its acceptance of this Agreement and agrees to be bound by the terms and conditions of this Agreement. Cover Page _Order Form_ **Cloud Service:** Zep is a cloud-based platform-as-a-service that offers fast, scalable, privacy-compliant building blocks for Generative AI apps. **Subscription Start Date:** The Effective Date **Subscription Period:** 1 month(s) **Non-Renewal Notice Period:** At least 30 days before the end of the current Subscription Period. **Cloud Service Fees:** Section 5.2 of the Standard Terms is replaced with: Certain parts of the Product have different pricing plans, which are available at Provider's [pricing page](https://www.getzep.com/pricing). Within the Payment Period, Customer will pay Provider fees based on the Product tier selected at the time of account creation and Customer's usage per Subscription Period. Provider may update Product pricing by giving at least 30 days notice to Customer (including by email or notification within the Product), and the change will apply in the next Subscription Period. **Payment Period:** 5 day(s) from the last day of the Subscription Period **Invoice Period:** Monthly _Key Terms_ **Customer:** The company or person who accesses or uses the Product. If the person accepting this Agreement is doing so on behalf of a company, all use of the word "Customer" in the Agreement will mean that company. **Provider:** Zep Software, Inc. **Effective Date:** The date Customer first accepts this Agreement. **Covered Claims:** **Provider Covered Claims:** Any action, proceeding, or claim that the Cloud Service, when used by Customer according to the terms of the Agreement, violates, misappropriates, or otherwise infringes upon anyone else's intellectual property or other proprietary rights. **Customer Covered Claims:** Any action, proceeding, or claim that (1) the Customer Content, when used according to the terms of the Agreement, violates, misappropriates, or otherwise infringes upon anyone else's intellectual property or other proprietary rights; or (2) results from Customer's breach or alleged breach of Section 2.1 (Restrictions on Customer). **General Cap Amount:** The fees paid or payable by Customer to provider in the 12 month period immediately before the claim **Governing Law:** The laws of the State of Delaware **Chosen Courts:** The state or federal courts located in Delaware **Notice Address:** For Provider: notices@getzep.com For Customer: The main email address on Customer's account _Changes to the Standard Terms_ **Publicity Rights:** Modifying Section 14.7 of the Standard Terms, Provider may identify Customer and use Customer's logo and trademarks on Provider's website and in marketing materials to identify Customer as a user of the Product. Customer hereby grants Provider a non-exclusive, royalty-free license to do so in connection with any marketing, promotion, or advertising of Provider or the Product during the length of the Agreement. **Benchmarking Restrictions:** Modifying Section 2.1 (Restrictions on Customer) of the Standard Terms, insert as clause (xi): **(xi)** conduct any performance testing, benchmarking, or comparative analysis of the Cloud Service for public or third-party disclosure without Provider’s prior written consent; provided that Customer may conduct such testing and benchmarking solely for its own internal evaluation purposes. **Survival of Benchmarking Restrictions:** Modifying Section 6.5 (Survival) of the Standard Terms, update sub-section 1 to include “Section 2.1(xi)” immediately after “Section 2.1 (Restrictions on Customer)” and update sub-section 2 to continue applying Section 2.1(xi) to retained Confidential Information: **6.5 Survival.** 1. The following sections will survive expiration or termination of the Agreement: Section 1.6 (Feedback and Usage Data), Section 2.1 (Restrictions on Customer), **Section 2.1(xi) (Benchmarking Restrictions)**, Section 5 (Payment & Taxes) … 2. Each Recipient may retain Discloser’s Confidential Information … in which case Section 4 (Privacy & Security), Section 12 (Confidentiality), **and Section 2.1(xi) (Benchmarking Restrictions)** will continue to apply to retained Confidential Information. --- ## Website terms of use **Source:** https://www.getzep.com/legal/website-terms/ [**We're hiring!** Come build with us →](/careers/) **Version 1.0** **Last revised on: January 27, 2024** The website located at getzep.com (the "**Site**") is a copyrighted work belonging to Zep Software, Inc. ("**Company**", "**us**", "**our**", and "**we**"). Certain features of the Site may be subject to additional guidelines, terms, or rules, which will be posted on the Site in connection with such features. All such additional terms, guidelines, and rules are incorporated by reference into these Terms. These Terms of Use (these "**Terms**") set forth the legally binding terms and conditions that govern your use of the Site. By accessing or using the Site, you are accepting these Terms (on behalf of yourself or the entity that you represent), and you represent and warrant that you have the right, authority, and capacity to enter into these Terms (on behalf of yourself or the entity that you represent). you may not access or use the Site or accept the Terms if you are not at least 18 years old. If you do not agree with all of the provisions of these Terms, do not access and/or use the Site. **PLEASE BE AWARE THAT SECTION 8.2 CONTAINS PROVISIONS GOVERNING HOW TO RESOLVE DISPUTES BETWEEN YOU AND COMPANY. AMONG OTHER THINGS, SECTION 8.2 INCLUDES AN AGREEMENT TO ARBITRATE WHICH REQUIRES, WITH LIMITED EXCEPTIONS, THAT ALL DISPUTES BETWEEN YOU AND US SHALL BE RESOLVED BY BINDING AND FINAL ARBITRATION. SECTION 8.2 ALSO CONTAINS A CLASS ACTION AND JURY TRIAL WAIVER. PLEASE READ SECTION 8.2 CAREFULLY.** **UNLESS YOU OPT OUT OF THE AGREEMENT TO ARBITRATE WITHIN 30 DAYS: (1) YOU WILL ONLY BE PERMITTED TO PURSUE DISPUTES OR CLAIMS AND SEEK RELIEF AGAINST US ON AN INDIVIDUAL BASIS, NOT AS A PLAINTIFF OR CLASS MEMBER IN ANY CLASS OR REPRESENTATIVE ACTION OR PROCEEDING AND YOU WAIVE YOUR RIGHT TO PARTICIPATE IN A CLASS ACTION LAWSUIT OR CLASS-WIDE ARBITRATION; AND (2) YOU ARE WAIVING YOUR RIGHT TO PURSUE DISPUTES OR CLAIMS AND SEEK RELIEF IN A COURT OF LAW AND TO HAVE A JURY TRIAL.** 1. **Accounts** 1.1. **Account Creation.** In order to use certain features of the Site, you must register for an account ("**Account**") and provide certain information about yourself as prompted by the account registration form. You represent and warrant that: (a) all required registration information you submit is truthful and accurate; (b) you will maintain the accuracy of such information. You may delete your Account at any time, for any reason, by following the instructions on the Site. Company may suspend or terminate your Account in accordance with Section 7. 1.2. **Account Responsibilities.** You are responsible for maintaining the confidentiality of your Account login information and are fully responsible for all activities that occur under your Account. You agree to immediately notify Company of any unauthorized use, or suspected unauthorized use of your Account or any other breach of security. Company cannot and will not be liable for any loss or damage arising from your failure to comply with the above requirements. 2. **Access to the Site** 2.1. **License.** Subject to these Terms, Company grants you a non-transferable, non-exclusive, revocable, limited license to use and access the Site solely for your own personal, noncommercial use. 2.2. **Certain Restrictions.** The rights granted to you in these Terms are subject to the following restrictions: (a) you shall not license, sell, rent, lease, transfer, assign, distribute, host, or otherwise commercially exploit the Site, whether in whole or in part, or any content displayed on the Site; (b) you shall not modify, make derivative works of, disassemble, reverse compile or reverse engineer any part of the Site; (c) you shall not access the Site in order to build a similar or competitive website, product, or service; and (d) except as expressly stated herein, no part of the Site may be copied, reproduced, distributed, republished, downloaded, displayed, posted or transmitted in any form or by any means. Unless otherwise indicated, any future release, update, or other addition to functionality of the Site shall be subject to these Terms. All copyright and other proprietary notices on the Site (or on any content displayed on the Site) must be retained on all copies thereof. 2.3. **Modification.** Company reserves the right, at any time, to modify, suspend, or discontinue the Site (in whole or in part) with or without notice to you. You agree that Company will not be liable to you or to any third party for any modification, suspension, or discontinuation of the Site or any part thereof. 2.4. **No Support or Maintenance.** You acknowledge and agree that Company will have no obligation to provide you with any support or maintenance in connection with the Site. 2.5. **Ownership.** You acknowledge that all the intellectual property rights, including copyrights, patents, trade marks, and trade secrets, in the Site and its content are owned by Company or Company's suppliers. Neither these Terms (nor your access to the Site) transfers to you or any third party any rights, title or interest in or to such intellectual property rights, except for the limited access rights expressly set forth in Section 2.1. Company and its suppliers reserve all rights not granted in these Terms. There are no implied licenses granted under these Terms. 2.6. **Feedback.** If you provide Company with any feedback or suggestions regarding the Site ("**Feedback**"), you hereby assign to Company all rights in such Feedback and agree that Company shall have the right to use and fully exploit such Feedback and related information in any manner it deems appropriate. Company will treat any Feedback you provide to Company as non-confidential and non-proprietary. You agree that you will not submit to Company any information or ideas that you consider to be confidential or proprietary. 3. **Indemnification.** You agree to indemnify and hold Company (and its officers, employees, and agents) harmless, including costs and attorneys' fees, from any claim or demand made by any third party due to or arising out of (a) your use of the Site, (b) your violation of these Terms or (c) your violation of applicable laws or regulations. Company reserves the right, at your expense, to assume the exclusive defense and control of any matter for which you are required to indemnify us, and you agree to cooperate with our defense of these claims. You agree not to settle any matter without the prior written consent of Company. Company will use reasonable efforts to notify you of any such claim, action or proceeding upon becoming aware of it. 4. **Third-Party Links & Ads; Other Users** 4.1. **Third-Party Links & Ads.** The Site may contain links to third-party websites and services, and/or display advertisements for third parties (collectively, "**Third-Party Links & Ads**"). Such Third-Party Links & Ads are not under the control of Company, and Company is not responsible for any Third-Party Links & Ads. Company provides access to these Third-Party Links & Ads only as a convenience to you, and does not review, approve, monitor, endorse, warrant, or make any representations with respect to Third-Party Links & Ads. You use all Third-Party Links & Ads at your own risk, and should apply a suitable level of caution and discretion in doing so. When you click on any of the Third-Party Links & Ads, the applicable third party's terms and policies apply, including the third party's privacy and data gathering practices. You should make whatever investigation you feel necessary or appropriate before proceeding with any transaction in connection with such Third-Party Links & Ads. 4.2. **Other Users.** Your interactions with other Site users are solely between you and such users. You agree that Company will not be responsible for any loss or damage incurred as the result of any such interactions. If there is a dispute between you and any Site user, we are under no obligation to become involved. 4.3.. **Release.** You hereby release and forever discharge Company (and our officers, employees, agents, successors, and assigns) from, and hereby waive and relinquish, each and every past, present and future dispute, claim, controversy, demand, right, obligation, liability, action and cause of action of every kind and nature (including personal injuries, death, and property damage), that has arisen or arises directly or indirectly out of, or that relates directly or indirectly to, the Site (including any interactions with, or act or omission of, other Site users or any Third-Party Links & Ads). IF YOU ARE A CALIFORNIA RESIDENT, YOU HEREBY WAIVE CALIFORNIA CIVIL CODE SECTION 1542 IN CONNECTION WITH THE FOREGOING, WHICH STATES: "A GENERAL RELEASE DOES NOT EXTEND TO CLAIMS WHICH THE CREDITOR OR RELEASING PARTY DOES NOT KNOW OR SUSPECT TO EXIST IN HIS OR HER FAVOR AT THE TIME OF EXECUTING THE RELEASE, WHICH IF KNOWN BY HIM OR HER MUST HAVE MATERIALLY AFFECTED HIS OR HER SETTLEMENT WITH THE DEBTOR OR RELEASED PARTY." 5. **Disclaimers** THE SITE IS PROVIDED ON AN "AS-IS" AND "AS AVAILABLE" BASIS, AND COMPANY (AND OUR SUPPLIERS) EXPRESSLY DISCLAIM ANY AND ALL WARRANTIES AND CONDITIONS OF ANY KIND, WHETHER EXPRESS, IMPLIED, OR STATUTORY, INCLUDING ALL WARRANTIES OR CONDITIONS OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE, QUIET ENJOYMENT, ACCURACY, OR NON-INFRINGEMENT. WE (AND OUR SUPPLIERS) MAKE NO WARRANTY THAT THE SITE WILL MEET YOUR REQUIREMENTS, WILL BE AVAILABLE ON AN UNINTERRUPTED, TIMELY, SECURE, OR ERROR-FREE BASIS, OR WILL BE ACCURATE, RELIABLE, FREE OF VIRUSES OR OTHER HARMFUL CODE, COMPLETE, LEGAL, OR SAFE. IF APPLICABLE LAW REQUIRES ANY WARRANTIES WITH RESPECT TO THE SITE, ALL SUCH WARRANTIES ARE LIMITED IN DURATION TO 90 DAYS FROM THE DATE OF FIRST USE. SOME JURISDICTIONS DO NOT ALLOW THE EXCLUSION OF IMPLIED WARRANTIES, SO THE ABOVE EXCLUSION MAY NOT APPLY TO YOU. SOME JURISDICTIONS DO NOT ALLOW LIMITATIONS ON HOW LONG AN IMPLIED WARRANTY LASTS, SO THE ABOVE LIMITATION MAY NOT APPLY TO YOU. 6. **Limitation on Liability** TO THE MAXIMUM EXTENT PERMITTED BY LAW, IN NO EVENT SHALL COMPANY (OR OUR SUPPLIERS) BE LIABLE TO YOU OR ANY THIRD PARTY FOR ANY LOST PROFITS, LOST DATA, COSTS OF PROCUREMENT OF SUBSTITUTE PRODUCTS, OR ANY INDIRECT, CONSEQUENTIAL, EXEMPLARY, INCIDENTAL, SPECIAL OR PUNITIVE DAMAGES ARISING FROM OR RELATING TO THESE TERMS OR YOUR USE OF, OR INABILITY TO USE, THE SITE, EVEN IF COMPANY HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGES. ACCESS TO, AND USE OF, THE SITE IS AT YOUR OWN DISCRETION AND RISK, AND YOU WILL BE SOLELY RESPONSIBLE FOR ANY DAMAGE TO YOUR DEVICE OR COMPUTER SYSTEM, OR LOSS OF DATA RESULTING THEREFROM. TO THE MAXIMUM EXTENT PERMITTED BY LAW, NOTWITHSTANDING ANYTHING TO THE CONTRARY CONTAINED HEREIN, OUR LIABILITY TO YOU FOR ANY DAMAGES ARISING FROM OR RELATED TO THESE TERMS (FOR ANY CAUSE WHATSOEVER AND REGARDLESS OF THE FORM OF THE ACTION), WILL AT ALL TIMES BE LIMITED TO A MAXIMUM OF FIFTY US DOLLARS. THE EXISTENCE OF MORE THAN ONE CLAIM WILL NOT ENLARGE THIS LIMIT. YOU AGREE THAT OUR SUPPLIERS WILL HAVE NO LIABILITY OF ANY KIND ARISING FROM OR RELATING TO THESE TERMS. SOME JURISDICTIONS DO NOT ALLOW THE LIMITATION OR EXCLUSION OF LIABILITY FOR INCIDENTAL OR CONSEQUENTIAL DAMAGES, SO THE ABOVE LIMITATION OR EXCLUSION MAY NOT APPLY TO YOU. 7. **Term and Termination.** Subject to this Section, these Terms will remain in full force and effect while you use the Site. We may suspend or terminate your rights to use the Site (including your Account) at any time for any reason at our sole discretion, including for any use of the Site in violation of these Terms. Upon termination of your rights under these Terms, your Account and right to access and use the Site will terminate immediately. Company will not have any liability whatsoever to you for any termination of your rights under these Terms, including for termination of your Account. Even after your rights under these Terms are terminated, the following provisions of these Terms will remain in effect: Sections 2.2 through 2.6 and Sections 3 through 8. 8. **General** 8.1. **Changes.** These Terms are subject to occasional revision, and if we make any substantial changes, we may notify you by sending you an e-mail to the last e-mail address you provided to us (if any), and/or by prominently posting notice of the changes on our Site. You are responsible for providing us with your most current e-mail address. In the event that the last e-mail address that you have provided us is not valid, or for any reason is not capable of delivering to you the notice described above, our dispatch of the e-mail containing such notice will nonetheless constitute effective notice of the changes described in the notice. Continued use of our Site following notice of such changes shall indicate your acknowledgement of such changes and agreement to be bound by the terms and conditions of such changes. 8.2. **Dispute Resolution.** Please read the following arbitration agreement in this Section (the "**Arbitration Agreement**") carefully.  It requires you to arbitrate disputes with Company, its parent companies, subsidiaries, affiliates, successors and assigns and all of their respective officers, directors, employees, agents, and representatives (collectively, the "**Company Parties**") and limits the manner in which you can seek relief from the Company Parties (a) **Applicability of Arbitration Agreement** You agree that any dispute between you and any of the Company Parties relating in any way to the Site, the services offered on the Site (the "**Services**") or these Terms will be resolved by binding arbitration, rather than in court, except that (1) you and the Company Parties may assert individualized claims in small claims court if the claims qualify, remain in such court and advance solely on an individual, non-class basis; and (2) you or the Company Parties may seek equitable relief in court for infringement or other misuse of intellectual property rights (such as trademarks, trade dress, domain names, trade secrets, copyrights, and patents). **This Arbitration Agreement shall survive the expiration or termination of these Terms and shall apply, without limitation, to all claims that arose or were asserted before you agreed to these Terms (in accordance with the preamble) or any prior version of these Terms.** This Arbitration Agreement does not preclude you from bringing issues to the attention of federal, state or local agencies. Such agencies can, if the law allows, seek relief against the Company Parties on your behalf. For purposes of this Arbitration Agreement, "**Dispute**" will also include disputes that arose or involve facts occurring before the existence of this or any prior versions of the Agreement as well as claims that may arise after the termination of these Terms. (b) **Informal Dispute Resolution.** There might be instances when a Dispute arises between you and Company. If that occurs, Company is committed to working with you to reach a reasonable resolution. You and Company agree that good faith informal efforts to resolve Disputes can result in a prompt, low‐cost and mutually beneficial outcome. You and Company therefore agree that before either party commences arbitration against the other (or initiates an action in small claims court if a party so elects), we will personally meet and confer telephonically or via videoconference, in a good faith effort to resolve informally any Dispute covered by this Arbitration Agreement ("**Informal Dispute Resolution Conference**"). If you are represented by counsel, your counsel may participate in the conference, but you will also participate in the conference. The party initiating a Dispute must give notice to the other party in writing of its intent to initiate an Informal Dispute Resolution Conference ("**Notice**"), which shall occur within 45 days after the other party receives such Notice, unless an extension is mutually agreed upon by the parties. Notice to Company that you intend to initiate an Informal Dispute Resolution Conference should be sent by email to: info@getzep.com, or by regular mail to 2261 Market Street #5686 San Francisco, CA 94114. The Notice must include: (1) your name, telephone number, mailing address, e‐mail address associated with your account (if you have one); (2) the name, telephone number, mailing address and e‐mail address of your counsel, if any; and (3) a description of your Dispute. The Informal Dispute Resolution Conference shall be individualized such that a separate conference must be held each time either party initiates a Dispute, even if the same law firm or group of law firms represents multiple users in similar cases, unless all parties agree; multiple individuals initiating a Dispute cannot participate in the same Informal Dispute Resolution Conference unless all parties agree. In the time between a party receiving the Notice and the Informal Dispute Resolution Conference, nothing in this Arbitration Agreement shall prohibit the parties from engaging in informal communications to resolve the initiating party's Dispute. Engaging in the Informal Dispute Resolution Conference is a condition precedent and requirement that must be fulfilled before commencing arbitration. The statute of limitations and any filing fee deadlines shall be tolled while the parties engage in the Informal Dispute Resolution Conference process required by this section. (c) **Arbitration Rules and Forum.** These Terms evidence a transaction involving interstate commerce; and notwithstanding any other provision herein with respect to the applicable substantive law, the Federal Arbitration Act, 9 U.S.C. § 1 et seq., will govern the interpretation and enforcement of this Arbitration Agreement and any arbitration proceedings. If the Informal Dispute Resolution Process described above does not resolve satisfactorily within 60 days after receipt of your Notice, you and Company agree that either party shall have the right to finally resolve the Dispute through binding arbitration. The Federal Arbitration Act governs the interpretation and enforcement of this Arbitration Agreement. The arbitration will be conducted by JAMS, an established alternative dispute resolution provider. Disputes involving claims and counterclaims with an amount in controversy under $250,000, not inclusive of attorneys' fees and interest, shall be subject to JAMS' most current version of the Streamlined Arbitration Rules and procedures available at http://www.jamsadr.com/rules-streamlined-arbitration/; all other claims shall be subject to JAMS's most current version of the Comprehensive Arbitration Rules and Procedures, available at http://www.jamsadr.com/rules-comprehensive-arbitration/. JAMS's rules are also available at www.jamsadr.com or by calling JAMS at 800-352-5267. A party who wishes to initiate arbitration must provide the other party with a request for arbitration (the "**Request**"). The Request must include: (1) the name, telephone number, mailing address, e‐mail address of the party seeking arbitration and the account username (if applicable) as well as the email address associated with any applicable account; (2) a statement of the legal claims being asserted and the factual bases of those claims; (3) a description of the remedy sought and an accurate, good‐faith calculation of the amount in controversy in United States Dollars; (4) a statement certifying completion of the Informal Dispute Resolution process as described above; and (5) evidence that the requesting party has paid any necessary filing fees in connection with such arbitration. If the party requesting arbitration is represented by counsel, the Request shall also include counsel's name, telephone number, mailing address, and email address. Such counsel must also sign the Request. By signing the Request, counsel certifies to the best of counsel's knowledge, information, and belief, formed after an inquiry reasonable under the circumstances, that: (1) the Request is not being presented for any improper purpose, such as to harass, cause unnecessary delay, or needlessly increase the cost of dispute resolution; (2) the claims, defenses and other legal contentions are warranted by existing law or by a nonfrivolous argument for extending, modifying, or reversing existing law or for establishing new law; and (3) the factual and damages contentions have evidentiary support or, if specifically so identified, will likely have evidentiary support after a reasonable opportunity for further investigation or discovery. Unless you and Company otherwise agree, or the Batch Arbitration process discussed in Subsection 8.2(h) is triggered, the arbitration will be conducted in the county where you reside. Subject to the JAMS Rules, the arbitrator may direct a limited and reasonable exchange of information between the parties, consistent with the expedited nature of the arbitration. If the JAMS is not available to arbitrate, the parties will select an alternative arbitral forum. Your responsibility to pay any JAMS fees and costs will be solely as set forth in the applicable JAMS Rules. You and Company agree that all materials and documents exchanged during the arbitration proceedings shall be kept confidential and shall not be shared with anyone except the parties' attorneys, accountants, or business advisors, and then subject to the condition that they agree to keep all materials and documents exchanged during the arbitration proceedings confidential. (d) **Authority of Arbitrator.** The arbitrator shall have exclusive authority to resolve all disputes subject to arbitration hereunder including, without limitation, any dispute related to the interpretation, applicability, enforceability or formation of this Arbitration Agreement or any portion of the Arbitration Agreement, except for the following: (1) all Disputes arising out of or relating to the subsection entitled "Waiver of Class or Other Non-Individualized Relief," including any claim that all or part of the subsection entitled "Waiver of Class or Other Non-Individualized Relief" is unenforceable, illegal, void or voidable, or that such subsection entitled "Waiver of Class or Other Non-Individualized Relief" has been breached, shall be decided by a court of competent jurisdiction and not by an arbitrator; (2) except as expressly contemplated in the subsection entitled "Batch Arbitration," all Disputes about the payment of arbitration fees shall be decided only by a court of competent jurisdiction and not by an arbitrator; (3) all Disputes about whether either party has satisfied any condition precedent to arbitration shall be decided only by a court of competent jurisdiction and not by an arbitrator; and (4) all Disputes about which version of the Arbitration Agreement applies shall be decided only by a court of competent jurisdiction and not by an arbitrator. The arbitration proceeding will not be consolidated with any other matters or joined with any other cases or parties, except as expressly provided in the subsection entitled "Batch Arbitration." The arbitrator shall have the authority to grant motions dispositive of all or part of any claim or dispute. The arbitrator shall have the authority to award monetary damages and to grant any non-monetary remedy or relief available to an individual party under applicable law, the arbitral forum's rules, and these Terms (including the Arbitration Agreement). The arbitrator shall issue a written award and statement of decision describing the essential findings and conclusions on which any award (or decision not to render an award) is based, including the calculation of any damages awarded. The arbitrator shall follow the applicable law. The award of the arbitrator is final and binding upon you and us. Judgment on the arbitration award may be entered in any court having jurisdiction. (e) **Waiver of Jury Trial.** EXCEPT AS SPECIFIED in section 8.2(a) YOU AND THE COMPANY PARTIES HEREBY WAIVE ANY CONSTITUTIONAL AND STATUTORY RIGHTS TO SUE IN COURT AND HAVE A TRIAL IN FRONT OF A JUDGE OR A JURY. You and the Company Parties are instead electing that all covered claims and disputes shall be resolved exclusively by arbitration under this Arbitration Agreement, except as specified in Section 8.2(a) above. An arbitrator can award on an individual basis the same damages and relief as a court and must follow these Terms as a court would. However, there is no judge or jury in arbitration, and court review of an arbitration award is subject to very limited review. (f) **Waiver of Class or Other Non-Individualized Relief.**  YOU AND COMPANY AGREE THAT, EXCEPT AS SPECIFIED IN SUBSECTION 8.2(h) EACH OF US MAY BRING CLAIMS AGAINST THE OTHER ONLY ON AN INDIVIDUAL BASIS AND NOT ON A CLASS, REPRESENTATIVE, OR COLLECTIVE BASIS, AND THE PARTIES HEREBY WAIVE ALL RIGHTS TO HAVE ANY DISPUTE BE BROUGHT, HEARD, ADMINISTERED, RESOLVED, OR ARBITRATED ON A CLASS, COLLECTIVE, REPRESENTATIVE, OR MASS ACTION BASIS. ONLY INDIVIDUAL RELIEF IS AVAILABLE, AND DISPUTES OF MORE THAN ONE CUSTOMER OR USER CANNOT BE ARBITRATED OR CONSOLIDATED WITH THOSE OF ANY OTHER CUSTOMER OR USER. Subject to this Arbitration Agreement, the arbitrator may award declaratory or injunctive relief only in favor of the individual party seeking relief and only to the extent necessary to provide relief warranted by the party's individual claim. Nothing in this paragraph is intended to, nor shall it, affect the terms and conditions under the Subsection 8.2(h) entitled "Batch Arbitration." Notwithstanding anything to the contrary in this Arbitration Agreement, if a court decides by means of a final decision, not subject to any further appeal or recourse, that the limitations of this subsection, "Waiver of Class or Other Non-Individualized Relief," are invalid or unenforceable as to a particular claim or request for relief (such as a request for public injunctive relief), you and Company agree that that particular claim or request for relief (and only that particular claim or request for relief) shall be severed from the arbitration and may be litigated in the state or federal courts located in the State of California. All other Disputes shall be arbitrated or litigated in small claims court. This subsection does not prevent you or Company from participating in a class-wide settlement of claims. (g) **Attorneys' Fees and Costs.** The parties shall bear their own attorneys' fees and costs in arbitration unless the arbitrator finds that either the substance of the Dispute or the relief sought in the Request was frivolous or was brought for an improper purpose (as measured by the standards set forth in Federal Rule of Civil Procedure 11(b)). If you or Company need to invoke the authority of a court of competent jurisdiction to compel arbitration, then the party that obtains an order compelling arbitration in such action shall have the right to collect from the other party its reasonable costs, necessary disbursements, and reasonable attorneys' fees incurred in securing an order compelling arbitration. The prevailing party in any court action relating to whether either party has satisfied any condition precedent to arbitration, including the Informal Dispute Resolution Process, is entitled to recover their reasonable costs, necessary disbursements, and reasonable attorneys' fees and costs. (h) **Batch Arbitration.** To increase the efficiency of administration and resolution of arbitrations, you and Company agree that in the event that there are 100 or more individual Requests of a substantially similar nature filed against Company by or with the assistance of the same law firm, group of law firms, or organizations, within a 30 day period (or as soon as possible thereafter), the JAMS shall (1) administer the arbitration demands in batches of 100 Requests per batch (plus, to the extent there are less than 100 Requests left over after the batching described above, a final batch consisting of the remaining Requests); (2) appoint one arbitrator for each batch; and (3) provide for the resolution of each batch as a single consolidated arbitration with one set of filing and administrative fees due per side per batch, one procedural calendar, one hearing (if any) in a place to be determined by the arbitrator, and one final award ("**Batch Arbitration**"). All parties agree that Requests are of a "substantially similar nature" if they arise out of or relate to the same event or factual scenario and raise the same or similar legal issues and seek the same or similar relief. To the extent the parties disagree on the application of the Batch Arbitration process, the disagreeing party shall advise the JAMS, and the JAMS shall appoint a sole standing arbitrator to determine the applicability of the Batch Arbitration process ("**Administrative Arbitrator**"). In an effort to expedite resolution of any such dispute by the Administrative Arbitrator, the parties agree the Administrative Arbitrator may set forth such procedures as are necessary to resolve any disputes promptly. The Administrative Arbitrator's fees shall be paid by Company. You and Company agree to cooperate in good faith with the JAMS to implement the Batch Arbitration process including the payment of single filing and administrative fees for batches of Requests, as well as any steps to minimize the time and costs of arbitration, which may include: (1) the appointment of a discovery special master to assist the arbitrator in the resolution of discovery disputes; and (2) the adoption of an expedited calendar of the arbitration proceedings. This Batch Arbitration provision shall in no way be interpreted as authorizing a class, collective and/or mass arbitration or action of any kind, or arbitration involving joint or consolidated claims under any circumstances, except as expressly set forth in this provision. (i) **30-Day Right to Opt Out.**  You have the right to opt out of the provisions of this Arbitration Agreement by sending a timely written notice of your decision to opt out to the following address: 2261 Market Street #5686, San Francisco, CA 94114, or email to info@getzep.com, within 30 days after first becoming subject to this Arbitration Agreement. Your notice must include your name and address and a clear statement that you want to opt out of this Arbitration Agreement. If you opt out of this Arbitration Agreement, all other parts of these Terms will continue to apply to you. Opting out of this Arbitration Agreement has no effect on any other arbitration agreements that you may currently have with us, or may enter into in the future with us. (j) **Invalidity, Expiration.** Except as provided in the subsection entitled "Waiver of Class or Other Non-Individualized Relief", if any part or parts of this Arbitration Agreement are found under the law to be invalid or unenforceable, then such specific part or parts shall be of no force and effect and shall be severed and the remainder of the Arbitration Agreement shall continue in full force and effect. You further agree that any Dispute that you have with Company as detailed in this Arbitration Agreement must be initiated via arbitration within the applicable statute of limitation for that claim or controversy, or it will be forever time barred. Likewise, you agree that all applicable statutes of limitation will apply to such arbitration in the same manner as those statutes of limitation would apply in the applicable court of competent jurisdiction. (k)**Modification.** Notwithstanding any provision in these Terms to the contrary, we agree that if Company makes any future material change to this Arbitration Agreement, you may reject that change within 30 days of such change becoming effective by writing Company at the following address: 2261 Market Street #5686, San Francisco, CA 94114, or email to info@getzep.com. Unless you reject the change within 30 days of such change becoming effective by writing to Company in accordance with the foregoing, your continued use of the Site and/or Services, including the acceptance of products and services offered on the Site following the posting of changes to this Arbitration Agreement constitutes your acceptance of any such changes. Changes to this Arbitration Agreement do not provide you with a new opportunity to opt out of the Arbitration Agreement if you have previously agreed to a version of these Terms and did not validly opt out of arbitration. If you reject any change or update to this Arbitration Agreement, and you were bound by an existing agreement to arbitrate Disputes arising out of or relating in any way to your access to or use of the Services or of the Site, any communications you receive, any products sold or distributed through the Site, the Services, or these Terms, the provisions of this Arbitration Agreement as of the date you first accepted these Terms (or accepted any subsequent changes to these Terms) remain in full force and effect. Company will continue to honor any valid opt outs of the Arbitration Agreement that you made to a prior version of these Terms. 8.3. **Export.** The Site may be subject to U.S. export control laws and may be subject to export or import regulations in other countries. You agree not to export, reexport, or transfer, directly or indirectly, any U.S. technical data acquired from Company, or any products utilizing such data, in violation of the United States export laws or regulations. 8.4. **Disclosures.** Company is located at the address in Section 8.8. If you are a California resident, you may report complaints to the Complaint Assistance Unit of the Division of Consumer Product of the California Department of Consumer Affairs by contacting them in writing at 400 R Street, Sacramento, CA 95814, or by telephone at (800) 952-5210. 9.5. **Electronic Communications.** The communications between you and Company use electronic means, whether you use the Site or send us emails, or whether Company posts notices on the Site or communicates with you via email. For contractual purposes, you (a) consent to receive communications from Company in an electronic form; and (b) agree that all terms and conditions, agreements, notices, disclosures, and other communications that Company provides to you electronically satisfy any legal requirement that such communications would satisfy if it were be in a hardcopy writing. The foregoing does not affect your non-waivable rights. 8.6. **Entire Terms.** These Terms constitute the entire agreement between you and us regarding the use of the Site. Our failure to exercise or enforce any right or provision of these Terms shall not operate as a waiver of such right or provision. The section titles in these Terms are for convenience only and have no legal or contractual effect. The word "including" means "including without limitation". If any provision of these Terms is, for any reason, held to be invalid or unenforceable, the other provisions of these Terms will be unimpaired and the invalid or unenforceable provision will be deemed modified so that it is valid and enforceable to the maximum extent permitted by law. Your relationship to Company is that of an independent contractor, and neither party is an agent or partner of the other. These Terms, and your rights and obligations herein, may not be assigned, subcontracted, delegated, or otherwise transferred by you without Company's prior written consent, and any attempted assignment, subcontract, delegation, or transfer in violation of the foregoing will be null and void. Company may freely assign these Terms. The terms and conditions set forth in these Terms shall be binding upon assignees. 8.7. **Copyright/Trademark Information**. Copyright ©2024 Zep Software, Inc. All rights reserved. All trademarks, logos and service marks ("**Marks**") displayed on the Site are our property or the property of other third parties. You are not permitted to use these Marks without our prior written consent or the consent of such third party which may own the Marks. **Contact Information:** Daniel Chalef Address: 2261 Market Street #5686 San Francisco, CA 94114 --- ## Zep vs. Letta: A Letta Alternative for Memory **Source:** https://www.getzep.com/letta-alternative/ Zep vs. Letta Letta (formerly MemGPT) is an open-source framework for building stateful agents with self-editing memory. Zep is the Context Lake — enterprise infrastructure that manages, governs, and serves agent memory at scale on temporal context graphs. Key takeaways ## A framework decision, or a _memory_ decision - Letta ([letta.com](https://www.letta.com/)) is a stateful-agent _framework_ with self-editing memory; Zep is a dedicated [agent-memory](/ai-agents/what-is-agent-memory/) layer — the [Context Lake](/platform/context-lake/). - Zep is framework-agnostic, stores bi-temporal facts with provenance, and governs memory in the data layer (ABAC, retention, audit; SOC 2 Type II, HIPAA). - Benchmark-proven: 94.7% LoCoMo and 90.2% LongMemEval ([results](https://www.getzep.com/research/)). S&P Global Market Intelligence named Zep a likely de facto partner in the enterprise agent stack ([coverage](https://www.getzep.com/research/sp-global-market-intelligence-zep-coverage/)). The distinction ## A framework vs. a memory layer **What Letta is.** Letta grew out of the MemGPT research and is an open-source platform for building agents that manage their own memory — with memory “blocks” the agent can edit, and a framework for orchestrating stateful agents. Its center of gravity is the agent framework: how the agent reasons and updates its own context. **What Zep is.** Zep is a dedicated memory layer, not a framework. It ingests chat, business data, and documents; builds bi-temporal context graphs (via the open-source [Graphiti](/platform/graphiti/), running on [Konig](/platform/agent-knowledge-graph/), Zep's proprietary graph database service); and serves relevant, token-efficient context to any agent — built in any framework, or none. Memory is governed in the data layer (ABAC, retention, audit) and served at enterprise scale with sub-200ms p95 retrieval. R Robbie 2024-09-07 · 14:27 I only wear Adidas shoes. I love them! Facts - Robbie only wears Adidas shoes. - Robbie strongly favors Adidas shoes. soleworks.com /account/returns/SO-48219 Soleworks Return · Order #SO-48219 · Adidas Ultraboost 22 Reason for return Product fell apart Additional comments These Adidas fell apart after three weeks and I'm furious . I'll be buying Nike from now on. Facts - Robbie only wears Adidas shoes. - Robbie strongly favors Adidas shoes. - Robbie ’s Adidas shoes fell apart . - Robbie is returning their Adidas shoes. - Robbie is angry about their Adidas shoes. - Robbie intends to wear Nike shoes. How they compare ## Letta vs. Zep, side by side Letta Zep Primary role Stateful-agent framework with self-editing memory Dedicated agent-memory layer (Context Lake) Framework lock-in You build on Letta's agent model Works with any framework, or none Memory model Agent-managed memory blocks Bi-temporal context graph (facts + provenance + validity) Temporal reasoning No temporal graph — OS-style memory blocks the agent self-edits “What's true now / what was true then,” automatic fact invalidation Enterprise governance Framework-level; not the focus Data-layer ABAC, retention + legal hold, audit, BYOK/BYOC, SOC 2 Type II, HIPAA Benchmarks — 94.7% LoCoMo (155ms), 90.2% LongMemEval (162ms) Open source Yes Graphiti (the graph library) is open source When to choose ## Pick the tool that fits the problem Choose Letta when You want an opinionated framework for building agents that manage their own memory, and you're happy to adopt that agent model end-to-end. - You're choosing an agent framework, not just a memory layer - Self-editing memory blocks out of the box are the priority - Research and smaller projects, open-source-first Choose Zep when Memory is the hard part and you don't want it coupled to a single agent framework. - Temporal, provenance-tracked facts across many sources - Governance and deployment control for regulated environments - Proven retrieval performance at enterprise scale - S&P Global Market Intelligence named Zep a likely de facto partner in the enterprise agent stack Get started ## Add governed agent memory in _three lines_ of code FAQ ## Frequently asked questions ### Is Letta a memory layer or a framework? Primarily a stateful-agent framework with self-editing memory. Zep is a dedicated memory layer that any framework can call. ### Can I use Zep with my existing agent framework? Yes — Zep is framework-agnostic (LangGraph, custom, or none) and adds memory in three lines of code. ### Which is better for enterprise? For governed memory at scale across many agents and data sources, Zep is purpose-built for it; Letta is centered on the agent framework. Evaluate both against your governance and scale requirements. --- ## The Mem0 alternative — Zep agent memory built for production **Source:** https://www.getzep.com/mem0-alternative/ The mem0 alternative Zep vs Mem0: Higher accuracy. Faster retrieval. Fewer tokens. Plus the customization, governance, and scale that production agents need. [Start building](https://app.getzep.com/api/auth/register) Trusted by teams building production agents Benchmarks ## Zep: More accurate, far faster, more token efficient Zep and mem0 run on the same long-running memory benchmarks: LoCoMo (1,540 questions) and LongMemEval (500 questions). Same reader prompts. Same judge. The retrieval stack is what varies. ### _LoCoMo_ 1,540 questions · long-running memory Accuracy 94.7 % Zep accuracy Zep 94.7% mem0 91.6% Retrieval 35× faster at p50 · 87 ms vs 3,060 ms Zep 87 ms mem0 3,060 ms Context 17% smaller 5,760 vs 6,956 tokens Zep 5,760 tok mem0 6,956 tok **Zep wins every category:** multi-hop, temporal, open-domain, single-hop. ### _LongMemEval_ 500 questions · top-k retrieval Accuracy 90.2 % Zep accuracy Zep 90.2% mem0 90.4% top\_50 Retrieval 24× faster at p50 · 104 ms vs 2,470 ms Zep 104 ms mem0 2,470 ms Context 35% smaller 4,408 vs 6,787 tokens Zep 4,408 tok mem0 6,787 tok Same accuracy at comparable retrieval depth, on a third less context and an order of magnitude faster. [See the full methodology and results](/research/)Mem0 baseline numbers published by mem0 at [github.com/mem0ai/memory-benchmarks](https://github.com/mem0ai/memory-benchmarks). Beyond chat memory ## Mem0 stores facts from conversation. Zep ingests every source the agent _touches_. ### Any data source Chat, JSON, app events, documents, business data. Zep ingests through a single SDK and unifies it in one graph per subject — user, customer, team, or topic. ### Custom entities and edges Define the entity types and relationships your domain needs. Zep extracts and links them at ingest, with your schema enforced. ### Temporal at the data model Every fact carries timestamps for when it became valid and when it stopped being valid. Point-in-time queries follow from the model, not from custom logic built above it. ## Governed at the data layer Govern context across thousands of agents, users, and context sources. Access control ### Attribute-based access control Control what context agents can access and what they can do with it. Retention ### Retention policies Retention is policy-driven. Data expires on the schedule you set. Legal hold blocks deletion when compliance requires it. Audit ### Audit and API logs Detailed logs of every request and policy decision, ready for audit. Enterprise scale ## Built for enterprise _scale_. Zep is built on [the Context Lake](/platform/context-lake/) — millions of context graphs per deployment, served back to agents in milliseconds. In production at Fortune 500 deployments. ### Sub-200ms retrieval at scale Hot graphs held in memory, with vector and BM25 indexes alongside. One call, one ranked answer. ### SOC 2 Type II and HIPAA Compliance posture, attestation reports, and controls in the [Trust Center](https://trust.getzep.com). ### Deploy your way Managed cloud, your own keys, or fully inside your VPC. The trust boundary moves with the deployment. [See deployment options](/enterprise/#deployment) Recognition ## Agent memory _infrastructure_ for the enterprise. [Read the report](/analysts/sp-market-intelligence-report/) [ S&P Global Market Intelligence ### Zep tackles agent memory limitations through its temporal context graph. S&P Global Market Intelligence · April 2026 ](/analysts/sp-market-intelligence-report/) We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack. — Melissa Incera, S&P Global Market Intelligence When to choose ## A clear _fit_, both ways. Zep Choose Zep when you need: - Facts that change over time, with point-in-time queries - Business data integrated alongside chat — CRM, support, billing, events - Custom entities and relationships specific to your domain - Retrieval that holds at sub-200ms across millions of subjects - Enterprise deployment with SOC 2 Type II, HIPAA, and BYOC mem0 Stay with mem0 when: - Basic chat-history memory is enough - You don't need temporal validity or business data integration - You're prototyping and quick setup matters more than production characteristics Migration ## Migrate to _Zep_. 01 ### Data transfer Step-by-step migration of your existing memory data. 02 ### Schema mapping Map your structures to Zep's temporal graph format. 03 ### Engineering support Direct support from Zep's team during migration. 04 ### Performance tuning Configure retrieval depth, scopes, and budgets for your workload. [View the migration guide](https://help.getzep.com/mem0-to-zep) Get started ## Move to production agent _memory_. --- ## Konig — Zep's proprietary graph database service **Source:** https://www.getzep.com/platform/agent-knowledge-graph/ Zep's proprietary graph database service The data plane beneath Zep’s agent memory platform. Most graph databases hold one large graph. The Context Lake needs millions of small ones — mostly cold, all temporal, all governed. [Technical blog post](https://blog.getzep.com/why-we-built-a-graph-database-service-for-agent-memory/) Architecture ## Built for the workload, not _retrofitted_. Konig is shaped for the Context Lake workload from the data model up. Millions of graphs. Sparse activity per graph. High aggregate throughput. Governance applied independently to each. Every architectural decision — tiered storage, in-memory adjacency, native ABAC, bi-temporal edges — follows from the workload. ### API Write mutate Read search (BFS, PageRank, Semantic, BM25) · patterns · list / get Typed surface for mutations, traversals, graph algorithms, and lookups. From API · writes ### WAL append-only · replicated · ordered Mutations durably appended to the write-ahead log before acknowledgment. From API · reads ### Graph Primitives Algorithms BFS · PageRank · pattern matching · path analysis · temporal weighting Structure adjacency & CSR matrices · vector index · BM25 index Graph algorithms, traversals, and pattern matching run over compact in-memory structures. ### Governance ABAC · multi-tenant isolation · customer key encryption · retention policies · audit · provenance Native to the data layer, not a layer bolted on. Every read and write is policy-gated for access and provenance; retention runs across the data lifecycle. ### Tiered Data Layer Hot RAM microseconds Warm Local NVMe microseconds Cold Object Store milliseconds Hot graphs serve at memory speed. Inactive graphs are evicted to NVMe and object-store, and rehydrated in milliseconds. ### Durable Storage WAL · content · metadata · graphs · indexes Snapshots and WAL persist on a multi-AZ, highly durable object and document store. Konig — request path from typed API down to durable storage Scale ## Scaling in _constant time_. Retrieval latency holds near-constant as the graph count grows. Current production runs sustain thousands of mutations and queries per second across millions of graphs — p50 latency unchanged from a thousand graphs to a million. Retrieval latency vs. graph count production · last 30 days p50 retrieval p95 retrieval Tiered storage ## Storage that tracks _activity_. Three tiers. Hot graphs live in RAM at microsecond latency. Warm graphs sit on local NVMe. Cold graphs rest on object storage and rehydrate in milliseconds. Cost tracks active graphs, not total graphs. A deployment with one million graphs and one percent of them hot pays for one percent of the memory. Hot RAM **microseconds** · resident Adjacency & CSR · vector & BM25 indexes Warm Local NVMe **microseconds** · paged Recent inactives, ready to rehydrate Cold Object Store **milliseconds** · rehydrate Long-tail graphs · cost-optimized § 05 · Retrieval ### Unified retrieval. Vector similarity, BM25, graph traversal, and pattern matching run over the same data. One query, one ranked answer — no separate retrieval stack to stitch together. § 06 · Primitives ### Graph primitives in memory. Hot graphs are held as adjacency lists and CSR matrices. Cache-friendly, deterministic layout, ready for matrix operations. BFS, PageRank, pattern matching, path analysis, and temporal weighting run at microsecond latencies over those structures. § 07 · Governance & temporality ### Governance and temporality, native. ABAC, multi-tenant isolation, customer key encryption, retention policies, audit, and provenance are properties of the data layer, not a layer above it. Every read and write is policy-gated. Every edge carries four timestamps. Point-in-time queries, automatic invalidation, and temporal weighting follow from the data model. § 08 · Durability ### Durable by default. Mutations are appended to a replicated, ordered write-ahead log before acknowledgment. Snapshots, indexes, and content persist on multi-AZ durable storage. Point-in-time recovery across the lifecycle. Get started ## Talk to the team _building it_. --- ## The Context Lake — Enterprise infrastructure for agent memory **Source:** https://www.getzep.com/platform/context-lake/ [**We're hiring!** Come build with us →](/careers/) Enterprise scale The data-lake pattern, applied to agent memory. Across every user, every domain, every agent, at scale. Context Lake 1,402,891 active graphs USER user\_8a32e1f9 247 1,204 2m ORG customer\_acme\_co 89 412 14s AGENT agent\_voyager 1,820 9,330 now DOMAIN domain\_billing 64 218 5m USER user\_d72b40c1 186 731 1m ORG customer\_initech 142 603 38s AGENT agent\_atlas\_v3 2,447 10,580 12s DOMAIN domain\_pricing 31 127 8m USER user\_4cc1ee07 312 1,486 4m ORG customer\_pied\_piper 77 298 22s AGENT agent\_helios 1,108 5,242 45s DOMAIN domain\_routing 48 174 3m USER user\_91ab7d2e 208 942 18s ORG customer\_globex 104 481 2m AGENT agent\_orion\_v2 1,640 7,820 8s DOMAIN domain\_inventory 56 198 6m Access control Retention Provenance Audit Trusted By AI Teams ## Why agent memory needs a lake [Memory for a single agent](/product/agent-memory/) is a solved problem. Memory across every agent, every user, every business unit is a different problem. Scale, isolation, governance, and retrieval performance all break at the same time. The Context Lake is the layer that solves them together. One system of record for context across the enterprise, governed at the entity level, served in milliseconds. Across _every_ agent. Every user. Every source of data. Context _Lake_ one system of record across all of it ## Inside the Context Lake Millions of context graphs, one per user, customer, team, or topic. Each captures the chat, documents, events, and business data tied to that subject — and the relationships between them — structured temporally and served back to agents on demand. ### Agent Runtime LangChain · LlamaIndex · CrewAI · Google ADK · custom Any agent framework — or none. The Context Lake is invoked through a single SDK. ### Ingestion chat · JSON · documents · app events Raw signal arrives from any source the agent touches. ### Context Assembly context blocks · templates · token-efficient Relevant context is assembled on demand into token-efficient blocks. ### Graphiti entity extraction · relationships · ontology · invalidation Signal becomes a temporal context graph as new facts arrive and stale ones are invalidated. ### Retrieval sub-200ms · auto-optimized · provenance-linked · policy-filtered Selects what's relevant and what adds the most information within the token budget. ### Governance ABAC · multi-tenant isolation · customer key encryption · retention policies · audit · provenance Native to the data layer, not a layer bolted on. Every read and write is policy-gated for access and provenance; retention runs across the data lifecycle. ### _Konig_ entities · facts & edges · decision traces · episodes Temporal context graph with provenance — sub-200ms retrieval at scale. ## Built on Konig Konig is Zep’s proprietary graph database service — millions of governed graphs, served in milliseconds, with isolation and temporality native to the data model. [Learn about Konig](/platform/agent-knowledge-graph/) Context Lake · section A–A drawing 06 · engine hot · in-memory working set cold · object-store snapshots scale 1:∞ Context Lake · section A–A drawing 06 · engine hot · in-memory working set cold · object-store snapshots scale 1:∞ ## A new layer in your data stack The Context Lake runs alongside your data lake, not as a replacement. Different data, different consumers, different access patterns. Same governance rigor. Data ### Structured, quantitative Tables, transactions, logs, metrics. ### Unstructured, qualitative Conversations, documents, events, decisions. Query model ### SQL, batch analytics Optimized for aggregation and retrospective analysis. ### Graph traversal, _semantic_ Temporal context graphs with entity-aware retrieval. Latency ### Seconds-to-minutes Batch jobs, scheduled pipelines, dashboard refresh. ### Sub-200ms retrieval Real-time context at agent inference speed. Consumers ### Dashboards & ML BI tools, data scientists, reporting pipelines. ### Agents & assistants LLM-powered applications that need memory and context. Governance ### Row & column ACLs Table-level permissions, role-based access. ### Entity-level ABAC Attribute-based policies, retention rules, full audit trail. Recognition ## Agent memory _infrastructure_ for the enterprise. [Read the report](/analysts/sp-market-intelligence-report/) [ S&P Global Market Intelligence ### Zep tackles agent memory limitations through its temporal context graph. S&P Global Market Intelligence · April 2026 ](/analysts/sp-market-intelligence-report/) We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack. — Melissa Incera, S&P Global Market Intelligence ## Governed at the data layer Govern context across thousands of agents, users, and context sources. Access control ### Attribute-based access control Control what context agents can access and what they can do with it. Retention ### Retention policies Retention is policy-driven. Data expires on the schedule you set. Legal hold blocks deletion when compliance requires it. Audit ### Audit and API logs Detailed logs of every request and policy decision, ready for audit. ## Choose your deployment model The trust boundary moves with your deployment. Choose where compute, data, and keys live. [Learn more.](/enterprise/) Trust boundary · Zep Zep Cloud Compute Data Keys Managed ### Cloud Zep's managed service. No infrastructure to run. Start in minutes. - SOC 2 Type II - HIPAA BAA Trust boundary · split Zep Cloud Compute Data Your KMS Keys AWS · GCP · Azure BYOK ### Cloud + Your Own Keys Zep's managed service with your own encryption keys. You control the keys; data at rest is encrypted with them. - SOC 2 Type II - HIPAA BAA Trust boundary · You Your VPC Zep service Compute Data Keys BYOC ### Bring Your Own Cloud Zep deployed inside your VPC. Your network, your perimeter, your compliance boundary. [Security & compliance](https://trust.getzep.com) ## Bring the Context Lake to your agents --- ## Graphiti — Zep **Source:** https://www.getzep.com/platform/graphiti/ Open source Open source and originated by Zep. [](https://github.com/getzep/graphiti)[Read the docs](https://help.getzep.com/graphiti/) ## What _Graphiti_ does. Graphiti turns conversations, business data, and documents into **temporal Context Graphs** — entities, relationships, and the timeline they live on. As facts change, Graphiti invalidates the old ones. Retrieval combines vector, full-text, and graph traversal in one call. ## More accurate. Faster. _Fewer tokens._ Agent memory systems often trade one for another. Zep leads on all three. BENCHMARK · LOCOMO 94.7% Accuracy Retrieval latency 155 ms Context size 5,760 tokens BENCHMARK · LONGMEMEVAL 90.2% Accuracy Retrieval latency 162 ms Context size 4,408 tokens [See the methodology and full results](/research/) Why Graphiti ## What makes Graphiti _different_. ### Temporal by design Every edge carries timestamps for when a fact became valid, when it stopped being valid, when Graphiti learned about it, and when it learned it was no longer true. Point-in-time queries follow from the data model. ### Hybrid retrieval Vector similarity, full-text search, and graph traversal in one ranked answer. No LLM-in-the-loop reranking, no orchestration layer to maintain. Invalid works\_at("Acme Inc.") Valid works\_at("Globex Corp.") ### Dynamic updates New facts integrate immediately. Outdated facts are invalidated by temporal logic — preserved as history, removed from current state. Neo4j FalkorDB Neptune OpenAI Azure Gemini Anthropic ### Pluggable backends Runs on Neo4j, FalkorDB, and Amazon Neptune. Supports OpenAI, Azure OpenAI, Google Gemini, and Anthropic. Model Context Protocol ## MCP _Server_. Connect Graphiti directly to Claude, Cursor, and other MCP-compatible clients. Give your tools graph-backed memory without changing your workflow. [Learn more about MCP](https://help.getzep.com/graphiti/getting-started/mcp-server) Powered by Zep ## Graphiti is the Context Graph framework. _Zep is the Context Lake._ Millions of governed Context Graphs, served in milliseconds, on top of Graphiti and Konig, Zep’s proprietary graph database service. Open source Graphiti The framework. One Context Graph per subject. The data model, the temporal logic, the hybrid retrieval API. Run it locally. [Read the docs](https://help.getzep.com/graphiti/) Commercial Zep The Context Lake. Millions of governed Context Graphs, served at sub-second latency. SOC 2, HIPAA, BYOC. [Learn about the Context Lake](/platform/context-lake/) [Konig](/platform/agent-knowledge-graph/) Community ## Build with the _community_. ### Star the project Help us reach more developers building the future of AI agents. [Star on GitHub](https://github.com/getzep/graphiti) ### Read the docs Comprehensive guides and examples. [View documentation](https://help.getzep.com/graphiti/) Contributing ## We welcome _contributions_ from the community. Bug fixes, documentation improvements, new features — your input makes Graphiti better. [View contribution guidelines](https://github.com/getzep/graphiti/blob/main/CONTRIBUTING.md) - 01 Report issues and suggest features on GitHub - 02 Improve documentation and examples - 03 Submit pull requests for bug fixes and features - 04 Answer questions on GitHub [](https://github.com/getzep/graphiti) --- ## Pricing **Source:** https://www.getzep.com/pricing/ Plans for every stage Credit-based plans for teams of every size. Enterprise plans for [production agent memory](/product/agent-memory/) at scale. Enterprise ## Agent memory, at _enterprise_ scale. Production agent memory with the security, compliance, and deployment flexibility large teams require. Custom credits, negotiated rates, and a guaranteed SLA — priced to your workload. [View deployment options](/enterprise/) - **Custom** credits with negotiated rates - Guaranteed rate limits with SLA - Unlimited projects - Custom Memory MCP Server seats - SOC 2 Type II and HIPAA BAA - Audit and API logs kept for 1-year - Slack/Teams support and dedicated account manager Fast growing venture-capital funded startup? [Get Zep Enterprise at an emerging company price.](/emerging/) Deploy anywhere Cloud Your VPC Self-serve ## Start in minutes. Scale with credits. ### Flex $125 / month billed monthly 50,000 credits included. Then $25 / 10,000 credits. Includes - **50,000** [credits](#faq-credits) per monthAuto top-up at 20%. 30-day rollover. - **600** requests per minute - **5** projects - **5** Memory MCP Server seats - [10 custom entity & edge types](https://help.getzep.com/customizing-graph-structure#custom-entity-and-edge-types) - API logs (1 day) - Unlimited memories, retrieval & users ### Flex Plus $375 / month billed monthly 200,000 credits included. Then $75 / 40,000 credits. Includes - **200,000** [credits](#faq-credits) per monthAuto top-up at 20%. 60-day rollover. - **1,000** requests per minute - **10** projects - **15** Memory MCP Server seats - [Observations](https://help.getzep.com/observations) - [20 custom entity & edge types](https://help.getzep.com/customizing-graph-structure#custom-entity-and-edge-types) - [Custom extraction instructions](https://help.getzep.com/custom-instructions) - [Webhooks](https://help.getzep.com/webhooks) - Analytics - API logs (7 days) - Unlimited memories, retrieval & users Prototyping? [Start free](https://app.getzep.com/api/auth/register) with 10,000 credits a month. [Limits apply](#faq-free-limits). Trusted By AI Teams One unit. Every operation ## How _credits_ work. Credits are consumed based on the size of each Episode you send to Zep — a chat message, JSON payload, or block of text. Memory, retrieval, storage, and users are unmetered. 1 credit per Episode up to 350 bytes; +1 credit per additional 350 bytes (or part). ⅛ credit per webhook invocation, where available. 0 credits for retrieval, storage, threads, users, and graph storage. Estimate your monthly spend Recommended: Flex Episodes / month15,000 0 250k 500k 750k 1M Average Episode size700 bytes · 2 credits ea. 100 B 1 kB 2 kB 3 kB 3.5 kB Webhook invocations / month20,000 0 1M 2M 3M 5M Credits / month 33k 32,500 credits Estimated cost $125 / month On the Flex plan ## Compare plans All limits, features, and security baselines across Flex, Flex Plus, and Enterprise. Flex $125 / month Flex Plus $375 / month Enterprise Custom Usage Included credits / month 50,000 200,000 Custom Overage rate $25 / 10k credits $75 / 40k credits Negotiated Auto top-up at 20% 10k credits ($25) 40k credits ($75) Custom Credit rollover 30 days 60 days Custom Free trial credits 10,000 / mo 10,000 / mo Custom Limits Requests per minute 600 1,000 Guaranteed, custom Projects 5 10 Unlimited Memory MCP Server Seats 5 15 Custom Custom entity & edge types 10 20 20 Memories, retrieval & users Unlimited Unlimited Unlimited Features Context Graph + temporal memory Observations — Custom extraction instructions — Webhooks — Analytics — Security & support SOC 2 Type II — — HIPAA BAA — — API logs retention 1 day 7 days 1 year Audit logs — — DPA (EU customers) — — Support Community Priority Slack/Teams · dedicated AM Deployment Cloud Cloud [Cloud · Cloud + BYOK · BYOC](/enterprise/) ## Choose your deployment model The trust boundary moves with your deployment. Choose where compute, data, and keys live. [Learn more.](/enterprise/) Trust boundary · Zep Zep Cloud Compute Data Keys Managed ### Cloud Zep's managed service. No infrastructure to run. Start in minutes. - SOC 2 Type II - HIPAA BAA Trust boundary · split Zep Cloud Compute Data Your KMS Keys AWS · GCP · Azure BYOK ### Cloud + Your Own Keys Zep's managed service with your own encryption keys. You control the keys; data at rest is encrypted with them. - SOC 2 Type II - HIPAA BAA Trust boundary · You Your VPC Zep service Compute Data Keys BYOC ### Bring Your Own Cloud Zep deployed inside your VPC. Your network, your perimeter, your compliance boundary. [Security & compliance](https://trust.getzep.com) Recognition ## Agent memory _infrastructure_ for the enterprise. [Read the report](/analysts/sp-market-intelligence-report/) [ S&P Global Market Intelligence ### Zep tackles agent memory limitations through its temporal context graph. S&P Global Market Intelligence · April 2026 ](/analysts/sp-market-intelligence-report/) We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack. — Melissa Incera, S&P Global Market Intelligence FAQ ## Frequently asked _questions_. Quick answers on credits, Episodes, security, and what’s included on each plan. Don’t see your question? [Contact sales](#contact) How credits work Credits are consumed based on the size of each [Episode](#faq-episode) you send to Zep. Episodes up to **350 bytes** use **1 credit**; each additional 350 bytes (or part thereof) uses another credit. Where available, webhook invocations consume **1/8 of a credit** each. **Flex** and **Flex Plus** automatically top up your credits when your balance drops below **20%**. Flex adds **10,000 credits ($25)**; Flex Plus adds **40,000 credits ($75)**. **Flex** credits roll over for **30 days**; **Flex Plus** credits roll over for **60 days**. Free plan credits do not roll over. What is an Episode? An **Episode** is any single data object you send to Zep — a chat message, JSON payload, or block of text. Credit cost scales with Episode size. Episodes up to **350 bytes** use **1 credit**; each additional 350 bytes (or part thereof) uses another credit. A 640-byte Episode uses **2 credits**; a 1,200-byte Episode uses **4 credits**. Are we charged for ingestion or storage? You are charged for ingestion and processing of Episodes. You are not charged for storage of messages or data. How are rate limits calculated? Each plan tier has a base rate limit, with Free being the lowest. Free and Flex Plan customers may see rate limits lowered depending on service usage. Enterprise plans have committed, guaranteed rate limits. What are the Free plan limits? - 10,000 credits per month. No rollover or auto-topup. - 2 projects, 1 Memory MCP Server seat, 5 custom entity & edge types. - Variable rate limits, depending on service-wide load. - Lower priority Episode processing. - Feature availability and service levels may change over time. Is Zep SOC 2 Type II certified? Yes. Zep is SOC 2 Type II certified. Review controls and live compliance status on the [Trust Center](https://trust.getzep.com). Can Zep sign a HIPAA BAA? Yes. HIPAA BAAs are available with the Enterprise plan. We're an EU company. Will you sign a DPA? Yes. Zep signs Data Processing Agreements with EU customers and customers with EU presence. ## Talk to the _team_. Start in minutes, or talk to sales about Enterprise deployment, BYOC, and custom limits. --- ## Agent Memory — Zep **Source:** https://www.getzep.com/product/agent-memory/ Agent Memory Ingest chat, business data, and documents. Shape the graph to your domain. Retrieve relevant context in under 200ms. Chat R Robbie 2024-09-07 I only wear Adidas shoes. I love them! Business data soleworks.com /returns/SO-48219 Soleworks Return · Adidas Ultraboost 22 Reason for return Product fell apart These Adidas fell apart after three weeks. I’ll be buying Nike from now on. Facts Extracted · 3 - Robbie strongly favors Adidas shoes. - Robbie ’s Adidas Ultraboost 22 fell apart . - Robbie will buy Nike next. Entities Relationships Timeline Trusted By AI Teams ## Memory Built on Context Graphs Create memories from any source. Zep constructs the graph. Retrieve relevant, token-efficient context. Any Source Chat History Business Data User Interactions People, things, and [how they change](#memory-validity). Automated Context Assembly context\_response.py ``` Emily prefers cycling to jogging (Valid: 2024-11-14 — present) # Observations Emily's blood pressure ranges 5% lower after cycling workouts. ``` ## Memory that understands when things change When new information contradicts what’s in the graph, Zep invalidates the old fact. Your agent reasons with the latest decisions, traits, and behaviors. Old facts stay as history. Ask what’s true now, or what was true on any past date. R Robbie 2024-09-07 · 14:27 I only wear Adidas shoes. I love them! Facts - Robbie only wears Adidas shoes. - Robbie strongly favors Adidas shoes. soleworks.com /account/returns/SO-48219 Soleworks Return · Order #SO-48219 · Adidas Ultraboost 22 Reason for return Product fell apart Additional comments These Adidas fell apart after three weeks and I'm furious . I'll be buying Nike from now on. Facts - Robbie only wears Adidas shoes. - Robbie strongly favors Adidas shoes. - Robbie ’s Adidas shoes fell apart . - Robbie is returning their Adidas shoes. - Robbie is angry about their Adidas shoes. - Robbie intends to wear Nike shoes. ## More accurate. Faster. Fewer tokens. Agent memory systems often trade one for another. [Zep leads on all three.](/research/) LoCoMo 94.7 % accuracy Retrieval latency 155 ms Context size 5,760 tokens LongMemEval 90.2 % accuracy Retrieval latency 162 ms Context size 4,408 tokens [See the methodology and full results](/research/) ## Memory patterns become Observations Zep analyses the structure of the graph to surface _Observations_: patterns, recurrences, and co-occurrences in memory. Your agent gains a global perspective, beyond facts and summaries. Jane has upgraded within _two weeks_ of each of the last _three product launches_. Apr 12, 2025 Jane upgraded to Pro v3 . +9d after launch Aug 4, 2025 Jane upgraded to Pro v4 . +11d after launch Nov 19, 2025 Jane upgraded to Pro v5 . +6d after launch ## Smart Context Assembly Zep picks the most valuable memory for the task at hand. Facts, summaries, and Observations selected to fit your token budget. Smarter than vector search alone, packed into a prompt-ready block. [Read the docs](https://docs.getzep.com/retrieving-context) **8 candidates** · ranked for task Obs Jane upgrades within 2 weeks of each launch. Fact Joined Aug 2024. Fact Currently on Pro v4. Fact Account billing monthly. Sum Recent chats: power-user features. Sum Past tickets: rate limits. Obs Tickets pair with plan changes. Fact Last login 12h ago. Context block 1,847 / 2,000 Obs Jane upgrades within 2 weeks of each launch. Fact Currently on Pro v4. Sum Recent chats: power-user features. Obs Tickets pair with plan changes. ## Token-efficient Context Blocks Zep returns a prompt-ready Context Block — shaped by your template — to drop into your agent’s prompt. [Read the docs](https://docs.getzep.com/context-templates) Your template   {{ user\_summary }}   {{ observations }}   {{ facts }} Context block   Pro user. Upgrades after each launch.   — Upgrades within 2 wks of launch.   — Tickets follow plan changes (8/9).   — On Pro v4 , billed monthly.   — Usage above Pro median.   — Last ticket: rate limits. Powered by Open Source ## Built on Graphiti, the _open-source_ Context Graph framework. Graphiti is the Context Graph framework that builds Zep’s Context Graphs. Open source at the core. [Read the docs](https://help.getzep.com/graphiti/) Apache 2.0 · Originated by Zep ## Three Lines of Code Add memory to your agent in minutes. Works with any agent framework, or none. [Read the Quickstart](https://help.getzep.com/v3/quick-start-guide) quickstart.py ```python # Add messages and get context in one call response = client.thread.add_messages( thread_id=thread_id, messages=[Message(name="Jane" , role="user" , content="I'd like to upgrade my plan..." )], return_context=True , )   # Add business data to the user's graph client.graph.add( user_id=user_id, type ="json" , data=json.dumps({"event" : "plan_upgrade" , "to" : "pro" , "mrr" : 49 }), )   # Get relevant context user_context = client.thread.get_user_context(thread_id=thread_id) ``` ## Customize for your domain Zep adapts to your business through custom entity types and relationship models. These models enable precise recall of exactly the context your agents need, not generic conversations. sales\_entities.py ```python class Lead (EntityModel ): """Represents a sales lead or prospect.""" company_size = Field( description="startup, SMB, mid-market, enterprise" ) budget_range = Field( description="Budget discussed or indicated" ) decision_timeline = Field( description="Expected decision timeframe" ) ``` ## Start with memory built for production --- ## Research — Zep **Source:** https://www.getzep.com/research/ Research Results on LoCoMo and LongMemEval, two industry benchmarks for long-running agent memory. We designed Zep for three constraints together: _accuracy_, _retrieval latency_, and _token efficiency_. Production agents need all three. Every decision in the Zep architecture follows from that. How we build the graph. How we retrieve. How we assemble context. The benchmarks below measure all three: how accurate Zep is, how fast it retrieves, and how many tokens it returns. LoCoMo 94.7 % accuracy Retrieval latency 155 ms Context size 5,760 tokens LongMemEval 90.2 % accuracy Retrieval latency 162 ms Context size 4,408 tokens ## LongMemEval LongMemEval evaluates long-running memory across six question types, including temporal reasoning and multi-session recall. Accuracy 90.2% 451 / 500 correct Retrieval latency 104/ 162ms p50 / p95 Median context 4,408tokens per question, end-to-end ### Accuracy by question type 0  —  100% Single-session assistant 96.4% Single-session user 94.3% Knowledge update 93.6% Temporal reasoning 90.2% Single-session preference 90.0% Multi-session 83.5% ## LoCoMo LoCoMo evaluates memory over multi-session conversations across four question categories: multi-hop, temporal, open-domain, and single-hop. Accuracy 94.7% 1,459 / 1,540 correct Retrieval latency 87/ 155ms p50 / p95 Median context 5,760tokens per question, end-to-end ### Accuracy by question category 0  —  100% Single-hop — 646 / 670 96.4% Temporal — 311 / 325 95.6% Multi-hop — 304 / 323 94.0% Open-domain — 175 / 221 79.2% ## Auto search: _half the tokens,_ no tuning. The results above use multi-scope retrieval — five parallel searches across facts, entities, episodes, observations, and thread summaries, composed at the client. Zep also offers **auto search**: a single API call that retrieves across every scope, applies a cross-scope rerank, and packs the result into a character-bounded context block. No scope selection, no client-side composition. Run on LoCoMo, auto search delivers: Accuracy 86.5% single API call, no tuning Retrieval latency 115/ 173ms p50 / p95 Median context 2,680tokens ↓ 53% smaller A single call, no scope tuning, with a context block _roughly half the size._ [Read the auto search docs](https://help.getzep.com/searching-the-graph#auto-search) ## How the results follow from the architecture. 01 ### Latency Hot graphs are held in memory as adjacency lists and CSR matrices, with vector and BM25 indexes alongside. One query returns one ranked answer across every retrieval signal. [Konig](/platform/agent-knowledge-graph/) 02 ### Token efficiency The Context Block is shaped at retrieval, not generated by an LLM. Entities, relationships, and observations are ranked at query time and packed to fit the token budget. [Agent Memory](/product/agent-memory/) 03 ### Accuracy Temporal invalidation keeps facts current. Pattern matching surfaces Observations. The retrieval ranks the right things on the first call. Methodology ## Reproducibility notes. Reader: `gpt-5.4` (reasoning = medium). Judge: `gpt-5.4` with chain-of-thought grading. Multi-scope retrieval depth: 20 edges, 10 nodes, 10 episodes, 5 thread summaries, 5 observations, cross-encoder reranking. Auto search at `max_characters=10000`. Run on 1,540 LoCoMo questions and 500 LongMemEval questions. 0 failed tests on either benchmark. ## Run Zep on _your own data._ [Start Building](https://app.getzep.com/api/auth/register) --- ## S&P Global Market Intelligence: Zep Tackles Agent Memory Limitations Through Its Temporal Context Graph **Source:** https://www.getzep.com/research/sp-global-market-intelligence-zep-coverage/ ## Key takeaways - S&P Global Market Intelligence positions Zep as a de facto partner in the enterprise agent stack — and a likely acquisition target as the ecosystem consolidates. - Zep’s differentiation is enterprise focus plus bi-temporal context graphs that integrate raw (episodic) and derived (semantic) memory and track how facts evolve over time. - On long, multisession benchmarks, Zep delivers meaningful accuracy and latency gains over full-context baselines while reducing token consumption. - Memory is the top capability enterprises expect from their agents (46.9% per S&P Global Market Intelligence’s Voice of the Enterprise data). ## Opening passage > “As enterprises advance along the agentic AI maturity curve, many are discovering how challenging it is to deliver reliably context-aware agents. Our Voice of the Enterprise data shows that memory is the top capability organizations expect from their agents (cited by 46.9% of respondents), highlighting how expectations are rising quickly, even as the underlying capabilities remain in early stages. Industry efforts have centered on traditional retrieval augmented generation and bolt-on memory. However, these approaches struggle with provenance, governance and temporality. Zep Software addresses these gaps by serving as the context layer for agents, unifying long-term memory, state and retrieval within a temporally aware graph. The result is fast context assembly with full preservation of decision provenance and access controls.” — Melissa Incera, _S&P Global Market Intelligence (451 Research) Market Insight Report_, April 10, 2026 ## Why this coverage matters Agent memory is the discipline of giving agents everything they need to know across time — about the user, the business, and the work being done. S&P Global Market Intelligence’s coverage initiation names the problem precisely: most agents today rely on retrieval augmented generation and bolt-on memory, and those approaches break down on provenance, governance, and temporality. Zep is the [Context Lake](/platform/context-lake/) for AI agents — the enterprise platform that manages, governs, and serves agent memory at scale. The report is an outside read on what we’ve built and why enterprises are deploying it. ## The Take > “Given that enabling stateful and context-aware agents has already become table stakes, it is striking how small the market around agentic memory remains. Only a handful of pure plays are tackling these challenges, while little by way of industry standards has emerged. We expect this subsector to accelerate rapidly, and Zep looks well positioned to capture that growth, provided it can scale accordingly. While the company is still very small today, we can easily see it becoming a de facto partner in this layer of the enterprise agent stack, if not a direct acquisition target, given the strong consolidation pressures across the ecosystem. Strategically, Zep's points of differentiation are its focus on the enterprise and its temporal understanding. On the latter, its graph integrates both raw (episodic) and derived (semantic) data, understanding both the temporal and emotional valence of it, and can therefore track how facts evolve over time. In benchmarks designed around longer multisession tasks, Zep's approach delivers meaningful improvements in accuracy and latency relative to full context baselines, while also reducing token consumption in long context scenarios.” — _The Take_, S&P Global Market Intelligence, April 10, 2026 ## What S&P Global Market Intelligence found 1. The agentic memory market is small but accelerating. > “Only a handful of pure plays are tackling these challenges, while little by way of industry standards has emerged. We expect this subsector to accelerate rapidly, and Zep looks well positioned to capture that growth.” 2. Zep is positioned to become a de facto partner in the enterprise agent stack. > “We can easily see it becoming a de facto partner in this layer of the enterprise agent stack, if not a direct acquisition target, given the strong consolidation pressures across the ecosystem.” 3. Differentiation: enterprise focus and temporal understanding. > “Strategically, Zep's points of differentiation are its focus on the enterprise and its temporal understanding... its graph integrates both raw (episodic) and derived (semantic) data, understanding both the temporal and emotional valence of it, and can therefore track how facts evolve over time.” 4. Benchmarked gains on long, multisession tasks. > “In benchmarks designed around longer multisession tasks, Zep's approach delivers meaningful improvements in accuracy and latency relative to full context baselines, while also reducing token consumption in long context scenarios.” 5. Memory is the top capability enterprises expect from their agents. > “Our Voice of the Enterprise data shows that memory is the top capability organizations expect from their agents (cited by 46.9% of respondents), highlighting how expectations are rising quickly, even as the underlying capabilities remain in early stages.” ## On the architecture > “Zep's base technology is Graphiti, an open-source, temporally aware knowledge-graph engine built specifically for agent memory. Zep structures this graph across three layers. The first is the episode subgraph, which captures raw messages, text and JSON as a complete, lossless source of truth. Above this sits a semantic entity subgraph, which extracts and organizes entities to create a richer, more navigable knowledge representation. The top layer is the community subgraph, which clusters related entities to provide broader contextual structure, and is responsible for chronological reasoning and for updating the truth-state when contradictions appear.” > “Crucially, Graphiti is bi-temporal in that it tracks both provenance (where information came from and when) and its validity over time, which ensures that outdated facts are deprecated as new information emerges. This enables agents to reason over a continuously evolving state, differentiating between what was true in the past and what is true now.” > “The company recently released an enterprise platform called Graphzilla \[Konig\], which serves as a graph engine supported by Zep's own vector database. A core differentiation from traditional, monolithic graph databases is the platform's 'lakehouse' (or 'context lake') concept architecture, which manages many medium-sized graphs (that could represent a specific user, team or project) rather than one single data store. To maintain fast query times, the system employs a hot graph memory management strategy, where only a percentage of active graphs are held in memory at any given time while the rest are continuously snapshotted and moved to cheap object storage.” — S&P Global Market Intelligence (451 Research), April 10, 2026 Note on naming: Konig, Zep’s proprietary graph database service, was developed internally under the codename Graphzilla. It powers the Context Lake — the enterprise system that manages, governs, and serves millions of context graphs as one. ## On the competitive landscape > “Competition for Zep is in a state of rapid evolution. Pure-plays are one piece of the puzzle, but approaches vary and many are still small. Mem0 is the highest-profile direct competitor and the best funded ($25 million raised in October 2025); however, it focuses more on the concept of portable memory and the personal assistant category rather than graphing complex enterprise relationships.” > “Zep's biggest competition at present will likely come from hyperscalers and agentic application providers that are memory primitives to make agents stateful. These would include Google Memory Bank in Vertex AI Agent Engine and Amazon Web Services AgentCore Memory... Crucially, these are platform-bound managed services bound to their ecosystems, whereas Zep stresses neutrality and graph-native semantics.” — S&P Global Market Intelligence (451 Research), April 10, 2026 ## SWOT analysis Strengths > “Zep AI's open-source foundation and research-driven approach provide a transparent, developer-first alternative to memory solutions offered by major cloud providers. The platform is uniquely engineered for performance-rich applications that demand real-time responsiveness. Its sophisticated temporal knowledge graphs offer a level of granular memory management that allows agents to accurately track how facts evolve over time, a critical differentiator for complex enterprise workflows.” Weaknesses > “Given the expanse of what Zep is attempting to do, the cost of running its system at enterprise scale is expensive. Zep is looking to optimize here, but we see cost sensitivity rising as it relates to AI and agents generally, which could push enterprises to find more economical competitive options.” Opportunities > “Our data corroborates that Zep is targeting a very real problem (enabling stateful agentic systems) and as there are not many in the space, there is huge opportunity to position itself as the neutral, multi-LLM memory layer for enterprises wary of hyperscaler lock-in, providing a consistent context strategy across diverse model providers like OpenAI, Anthropic and Meta.” Threats > “Zep's biggest threat will likely come from the large vendors going to market with full agentic stacks — hyperscalers and large model providers. These incumbents can leverage existing enterprise credits and deep ecosystem integrations to offer good enough memory solutions as a free or bundled feature, potentially commodifying Zep's core value proposition.” — S&P Global Market Intelligence (451 Research), April 10, 2026 ## About the analyst **[Melissa Incera](https://www.spglobal.com/marketintelligence/contributors/2321764/melissa-incera)** is a Senior Industry Analyst on the Data, AI & Analytics team within S&P Global Market Intelligence’s [451 Research](https://451research.com), covering applied AI and emerging enterprise AI infrastructure. **S&P Global Market Intelligence** is a division of S&P Global providing data, research, and analytics on companies, markets, and industries. The 451 Research team focuses on emerging technology and enterprise IT innovation. [spglobal.com/market-intelligence](https://www.spglobal.com/market-intelligence) ## Read the report Download the full S&P Global Market Intelligence coverage initiation, or schedule a conversation with our team about deploying agent memory at scale. [Download the full report (PDF)](/downloads/451Research_Reprint_ZepSoftware_10APR2026.pdf) ## Read next - [Read about the Context Lake →](/platform/context-lake/) - [Explore Graphiti, our open-source context graph framework →](/platform/graphiti/) - [See enterprise plans →](/enterprise/) This report, licensed to Zep Software, developed and as provided by S&P Global Energy (S&P), was published as part of S&P's syndicated market insight subscription service. It shall be owned in its entirety by S&P. This report is solely intended for use by the recipient and may not be reproduced or re-posted, in whole or in part, by the recipient without express permission from S&P. ©2026 by S&P Global Inc. All rights reserved. --- ## Buy vs. Build Agent Memory: A Decision Framework **Source:** https://www.getzep.com/resources/buy-vs-build-agent-memory/ [**We're hiring!** Come build with us →](/careers/) ## Key takeaways - The decision turns on one question: is [agent memory](/ai-agents/what-is-agent-memory/) your product, or plumbing beneath your product? - A production memory system has to do five things at once — ingestion, temporal correctness, budgeted retrieval, governance, and scale — and the cost lives in the interactions between them. - Build estimates go wrong because they're written against a chat-history buffer, not the system that buffer grows into. - Buy when enterprise requirements show up early; access control, audit, retention, and residency are the most expensive things to retrofit. Every agent that does more than answer one-shot questions needs [agent memory](/ai-agents/what-is-agent-memory/): what it knows across time about its users, the business, and the world it operates in. Without it, the agent starts from scratch every turn. With it, the agent can reason about the user and act on what it already knows. Many teams build memory themselves. It starts as a chat-history buffer. Then a vector store gets added. Then a layer of glue code holds the two together. The whole thing works in the demo, so it ships. This page is about what happens after the demo: when the build path stops working, and how to decide before it costs you a rewrite. Building your own is sometimes the right call, and the framework below should hold up in a code review and a budget meeting. ## What agent memory actually requires The build-vs-buy math goes wrong in the same place almost every time: the problem gets under-scoped. A buffer of recent messages looks like memory, so the estimate gets written against that. The estimate is for a different, smaller problem. A production memory system has to do five things at once. - **Ingestion and extraction.** A customer leaves a trail across more than chat: CRM records, support tickets, billing events, documents, product telemetry. All of it has to become structured facts the agent can query, unified per subject rather than scattered across stores. “The user switched from the Pro plan to Enterprise in March” is a fact with a subject, a relationship, and a timestamp. Pulling that out of free text and event streams reliably, across millions of subjects, is its own engineering problem. - **Temporal correctness.** Facts change. A user's job title, their preferences, their account tier — all of it moves. A memory system has to track what is true now and what was true at any past moment. That means [bi-temporal validity](/ai-agents/temporal-knowledge-graph/): the system records when a fact held in the world and when it learned the fact, and it invalidates the old fact when a new one arrives. Get this wrong and the agent acts on a preference the user abandoned three weeks ago. - **Retrieval that fits a token budget.** You cannot dump a user's entire history into the prompt. Retrieval has to pick the most valuable context for the task in front of the agent and fit it to a budget. Too little and the agent misses something it needed. Too much and cost climbs while accuracy drops. - **Governance.** The moment a second user exists, one user's data must be invisible to an agent acting for another. Add retention rules and audit logging, with provenance back to source. This is the part that turns a prototype into something you can put in front of an enterprise security review. - **Scale.** All of the above, for millions of subjects, with retrieval fast enough to sit in the request path. Sub-200ms under concurrent load, without the cost curve bending the wrong way. Each capability looks tractable on its own. The cost lives in the interactions. The hard part is making all of them hold together at once: temporal correctness that still respects access control while retrieval stays under 200ms. That is a much harder system than any single feature suggests. **Figure 1 — What a production memory system must do.** Five capabilities at once — ingestion, temporal correctness, retrieval in budget, governance, and scale — with the cost living in the interactions between them, not in any single feature. ## The build path: what you are actually signing up for The first version is a weekend of work: recent messages in a buffer, embeddings in a vector database, a similarity search at query time. For a prototype or a single-user tool, that is a reasonable place to stop. The trouble starts as usage grows, and it tends to arrive in a predictable order. First, **stale facts**. The vector store has no concept of time, so it retrieves a fact the user has since contradicted. The agent acts on it. Now you are writing logic to detect and supersede outdated facts — the temporal problem you skipped earlier, back to collect its debt. Then **context bloat**. As history grows, similarity search returns more loosely related chunks. The prompt gets longer and inference costs rise; answer quality slips because the signal is buried under the noise. You start building re-ranking and summarization to compensate. Then **access boundaries**. A customer asks the obvious security question: can one user's data reach another user's agent? Nothing in a shared vector index prevents it, and retrofitting entity-level access control into a system built without it is expensive and risky. Then the queries the vector store cannot answer at all — “which of this user's projects share a stakeholder with the account that just churned?” is a graph question, and cosine similarity does not answer it. The deeper cost is not any single fix. It is that you now own a system. Someone is on call for it. Someone maintains the eval harness, migrates it when the underlying models change, re-indexes when the schema moves, and debugs retrieval-quality regressions. That is a team working on memory infrastructure instead of the product you are actually trying to build. The initial build might be a few engineer-months; the maintenance is a standing tax with no end date. None of this means building is wrong. It means the build estimate has to include the second system — the one you grow into, not the buffer you start with. Weekend prototype Vector store Message buffer Then: stale facts \+ Temporal logic Vector store Message buffer Then: context bloat \+ Re-ranking & summaries Temporal logic Vector store Message buffer Then: access boundaries \+ Entity-level ACL Re-ranking & summaries Temporal logic Vector store Message buffer Now: a system you run \+ Graph store Entity-level ACL Re-ranking & summaries Temporal logic Vector store Message buffer v1 usage grows → **Figure 2 — What the build path grows into.** A weekend prototype of a message buffer plus a vector store accretes temporal logic, re-ranking, entity-level access control, and a graph store as usage grows — until what you maintain is a system you operate. ## The options in between Between hand-rolling memory and adopting a dedicated platform sits a real field of options. Three categories matter, and each is the right pick for some teams. - **Open source.** Two flavors matter here. mem0 is a memory store you self-host: a vector index and a key-value store, with a graph backend offered as a separate add-on. Facts are mutated in place, and you trigger and assemble retrieval yourself. [Graphiti](/platform/graphiti/), Zep's own open-source layer, goes further and builds a bi-temporal context graph with entity extraction and fact invalidation, plus a hybrid retrieval API — the most capable open option. Either way, the system around the store is yours to run: scale, multi-tenant isolation, governance, deployment, and compliance. - **Framework-bundled memory.** Build on an agent framework like Mastra and memory comes built in, and it is more than a scratchpad: working memory keeps a structured profile across threads, semantic recall pulls relevant past messages back by vector search, and background agents compress raw history into a dense, reflected log. The limit is the input. It works from the conversation alone — the same scope as mem0 and most memory layers — not the stream of touchpoints a customer leaves across the enterprise (CRM, support tickets, billing, documents, product events). Nor does it build a bi-temporal graph, so facts carry no validity windows and there are no point-in-time queries, and governance belongs to the framework rather than the memory layer. - **Hyperscaler primitives.** Amazon Bedrock [AgentCore Memory](/aws-agentcore-memory-alternative/) is the example: a managed service giving agents short-term session memory and long-term insights extracted asynchronously, inside the AWS and Bedrock environment. If your stack is committed to AWS, the native integration is the draw. The trade shows up in three places — lock-in (your agents' memory, among the stickiest data you own, is bound to one vendor's stack), governance (entity-level access control and retention sit outside the memory layer), and depth (an extraction-based store is shallower than a bi-temporal graph, with no way to ask what was true at a past moment or to trace a fact back to its source). Each option is good enough for the job it was built for. None of them combines bi-temporal facts, entity-level governance on every query, ingestion across chat and business systems, and sub-200ms retrieval at scale on a neutral stack. That combination is the job of a dedicated platform. Requirement Build from scratch mem0 Mastra AgentCore Graphiti Zep Ingests beyond chat (CRM, billing, docs, events) DIY Conversation only Conversation only Conversational + custom events Yes Yes Bi-temporal facts / point-in-time DIY No, mutated in place No No, extraction-based Yes Yes Entity-level governance (ABAC, retention, audit) DIY Account-level Outside memory layer Outside memory layer DIY Yes, in the data layer Sub-200ms at millions of subjects DIY Depends on setup Framework-bound Managed on AWS Depends on setup Yes, 155–162ms Neutral across model and cloud Yes Yes Yes No, AWS/Bedrock Yes Yes Who runs the surrounding system You You You AWS You Zep ## The buy path: what you get and what you give up Buying agent memory means the five capabilities above arrive as infrastructure rather than a roadmap. Ingestion, temporal facts, governed retrieval, and scale are someone else's maintenance burden. Zep is agent memory at enterprise scale. The [Context Lake](/platform/context-lake/) is how we believe it should be done: a governed system of context graphs that manages, governs, and serves everything an agent needs to know. It ingests every source the agent touches and unifies them into one context graph per subject — the conversation is one input among many. Zep runs the open-source [Graphiti](/platform/graphiti/) framework inside a managed system and serves the result through [Konig](/platform/agent-knowledge-graph/), its proprietary graph database service, with sub-200ms retrieval whether you have one context graph or millions. Every query is filtered by entity-level access control; facts carry validity windows and invalidate cleanly when they change; retrieval returns a prompt-ready context block that fits a token budget. SOC 2 Type II and HIPAA come with it, and it stays neutral across model and cloud — the opposite trade from a hyperscaler primitive. On the public benchmarks Zep reports 94.7% on LoCoMo at 155ms and 90.2% on LongMemEval at 162ms ([benchmark results](/research/)). The durable difference is architectural. What you give up is real, and worth stating plainly. You take on a dependency and a pricing relationship. You have less control over the internals than you would with code you wrote. And adopting any memory system is integration work, not a switch you flip. Three objections usually decide this. **Data control:** where does the data live, and who can touch it? Deployment models include fully managed, plus bring-your-own-key and bring-your-own-cloud for keeping data in your own environment. **Lock-in:** the graph-construction layer, Graphiti, is open source, so the foundation is not a black box you can never leave. **The “we're different” worry:** customization lives in how graphs are built and how context is assembled, not in forking the storage engine. ## How to decide One question decides it. Is agent memory your product, or plumbing beneath your product? **Build it yourself** when memory is the thing you sell, or a core differentiator you have to own end to end. Build when your requirements are narrow and stable: a single tenant, with no cross-user governance and a modest scale that is not going to move. Build when you have the team and the real appetite to carry the maintenance for years, not the launch. **Buy** when memory sits beneath the product rather than being the product. Buy when enterprise requirements show up early, because access control, audit, retention, and data residency are the capabilities that are most expensive to retrofit. Buy when time-to-production and keeping your engineers on your core product matter more than owning the internals. **Figure 3 — How to decide.** One question decides it: is agent memory your product, or plumbing beneath it? Build when it's the thing you sell; buy when enterprise requirements show up early; use open source when you have the appetite to operate the system for years. Take these questions to your team before you commit: - Is memory a differentiator we must own, or a dependency we can rent? - Will we need entity-level access control within 12 months? - What does it cost us, concretely, if the agent acts on a stale fact? - Who owns this system at 2am, 18 months from now? - If our first version does not scale, what is the rewrite worth? In the cost comparison, build is engineer-months plus a recurring maintenance tax; buy is a subscription plus integration effort. The line item that dominates is rarely the license. It is the maintenance. ## Where to start The right answer depends on whether memory is your product or your plumbing, and that is a question only your team can answer. If you want the open core and intend to run the surrounding system yourself, start with [Graphiti](/platform/graphiti/). If memory is plumbing you would rather not build and operate, [talk to us](/enterprise/). * * * _Related: [What is agent memory?](/ai-agents/what-is-agent-memory/) · [What is a Context Lake?](/platform/context-lake/) · [What is a temporal knowledge graph?](/ai-agents/temporal-knowledge-graph/) · [How to give an AI agent long-term memory](/ai-agents/how-to-give-ai-agents-long-term-memory/) · [AI agent memory guides](/ai-agents/)_ ## Frequently asked questions ### Should I build or buy agent memory? Build it when memory is the product you sell or a differentiator you must own end to end, your requirements are narrow and stable, and you have a team to carry the maintenance for years. Buy it when memory sits beneath the product, enterprise requirements (access control, audit, retention, residency) show up early, and time-to-production matters more than owning the internals. ### When does building your own agent memory stop working? Usually in a predictable order as usage grows: stale facts (the vector store has no concept of time), context bloat (similarity search returns more loosely related chunks), access boundaries (a shared index can't keep one user's data out of another's agent), and graph-shaped queries cosine similarity can't answer. The deeper cost is that you now own a system someone is on call for. ### What does a production agent memory system actually need to do? Five things at once: ingest and extract facts from every source (not just chat), track [bi-temporal](/ai-agents/temporal-knowledge-graph/) validity so it knows what's true now versus then, retrieve the most valuable context within a token budget, enforce governance (entity-level access control, retention, audit), and do all of it for millions of subjects with sub-200ms retrieval. The cost lives in the interactions between these, not in any single feature. ### Is open source like Graphiti or mem0 a middle path? Yes. [Graphiti](/platform/graphiti/) builds a bi-temporal context graph with entity extraction, fact invalidation, and hybrid retrieval — the most capable open option. mem0 is a self-hosted vector + key-value store with a graph add-on. Either way, the system around the store — scale, multi-tenant isolation, governance, and compliance — is yours to run. ### What's the real cost of building agent memory in-house? The initial build is a few engineer-months; the maintenance is a standing tax with no end date — an eval harness, model migrations, re-indexing on schema changes, and retrieval-quality debugging. In the cost comparison, build is engineer-months plus that recurring tax; buy is a subscription plus integration effort. The line item that dominates is rarely the license — it's the maintenance. ### Does buying agent memory lock me in? With Zep, the graph-construction layer is the open-source [Graphiti](/platform/graphiti/), so the foundation isn't a black box you can never leave, and it stays neutral across model and cloud. Deployment models include fully managed plus bring-your-own-key and bring-your-own-cloud for keeping data in your own environment — the opposite trade from a hyperscaler primitive bound to one vendor's stack. --- ## Context Engineering for AI Agents — Zep **Source:** https://www.getzep.com/solutions/context-engineering/ The practice The practice of designing, scoping, and managing the context AI agents need to reason, decide, and act. The discipline ## What is _context engineering?_ Context engineering is the discipline of getting the right information to an agent at the right time. Not the prompt. Not the model. The information layer underneath — what the agent knows about the user, the business, the history, and the moment it’s reasoning about. Every production agent does context engineering, whether intentionally or by accident. The choice is whether to design it. Context engineering covers four questions: - 01 What context exists Chat history, business data, documents, events, decisions. - 02 What context is relevant To this user, this task, this turn. - 03 How context is governed Who can see what, how long it's kept, where it traces back to. - 04 How context is delivered Assembled, ranked, packed into a budget the model can use. When teams do this well, agents reason with the user’s full history and the business’s current state. When they don’t, agents hallucinate, contradict, or fail to act on what’s already known. Category vs. practice ## Agent memory is the category. Context engineering is the _practice_. Agent memory is the _category_ of infrastructure that makes context engineering possible at scale. It’s what an agent knows across time — about the users it serves, the business it operates in, and the work it has done before. You can do agent memory badly: a file the agent writes to, a chat history buffer, a vector store that retrieves “similar” documents. You can do it well: a temporal graph that keeps facts current, governed at the entity level, served back to the agent at production latency. Context engineering is the practice. Agent memory is what the practice produces. Infrastructure ## The Context Lake is how Zep _delivers it_. The Context Lake is the infrastructure for agent memory at enterprise scale. Millions of context graphs, one per user, customer, team, or topic. Each captures the chat, documents, events, and business data tied to that subject, structured temporally and served back to agents on demand. 01 · Unified ### A single source of context Chat, documents, events, business data — all unified into one graph per subject. No more stitching together a vector store, a chat buffer, and a database every time the agent needs to reason. chat stream events kafka docs s3 biz data jdbc ↓ one graph per subject 02 · Governed ### Governance built in Attribute-based access control on every entity. Retention rules applied at ingest. Full audit. Context engineering at the enterprise level isn't possible without governance in the data layer. ABAC Retention Audit Provenance 03 · At scale ### Retrieval that holds at scale [Sub-200ms retrieval across millions of graphs](/platform/agent-knowledge-graph/). Vector, full-text, and graph traversal in one ranked answer. Context engineering ends at the moment the agent asks for context — the system has to be there. 155ms p95 · 1.4M graphs [Learn about the Context Lake](/platform/context-lake/) Why Zep ## Why teams choose Zep for _context engineering_. Reason 01 ### Multi-source by default Chat, business data, documents, events — ingested through a single SDK, unified in one graph per subject. Reason 02 ### Temporal at the data model Every fact carries timestamps for when it became valid and when it stopped being valid. Point-in-time queries follow from the model, not from custom logic above it. Reason 03 ### Governed at the data layer ABAC, retention, audit, and provenance are properties of the data model, not a layer bolted on. Every read and write is policy-gated. Recognition ## Agent memory _infrastructure_ for the enterprise. [Read the report](/analysts/sp-market-intelligence-report/) [ S&P Global Market Intelligence ### Zep tackles agent memory limitations through its temporal context graph. S&P Global Market Intelligence · April 2026 ](/analysts/sp-market-intelligence-report/) We can easily see Zep becoming a de facto partner in this layer of the enterprise agent stack. — Melissa Incera, S&P Global Market Intelligence FAQ ## Frequently _asked_. ### What's the difference between context engineering and prompt engineering? Prompt engineering is about how you instruct the model. Context engineering is about what the model sees alongside the instruction — the user history, business state, prior decisions, and current facts that shape the answer. Prompt engineering writes the question. Context engineering supplies the world the question is asked in. ### How is context engineering different from RAG? Retrieval-augmented generation is one technique inside context engineering. RAG retrieves documents from a corpus and adds them to the prompt. Context engineering covers everything else as well: user memory, business data, event streams, temporal validity, governance, and how the retrieved context is assembled and ranked. RAG is a tactic. Context engineering is the practice. ### Do I need a Context Lake to do context engineering? For a single-agent prototype, no. For production agents at enterprise scale, yes. Context engineering for thousands of agents across every user, customer, and team requires infrastructure that handles isolation, governance, retrieval performance, and temporal correctness together. That's what the Context Lake is built for. Get started ## Talk to the _team_. --- ## Supermemory Alternative — Agent Memory at Enterprise Scale **Source:** https://www.getzep.com/supermemory-alternative/ Zep vs. Supermemory Supermemory is built for conversational memory. Zep fuses every customer touchpoint across the enterprise — chat, CRM, support tickets, billing, documents, app events — into one governed context graph per subject, with bi-temporal facts and sub-200ms retrieval at enterprise scale. Key takeaways ## The conversation, or the whole _customer_ - Supermemory is built for conversational memory. Zep fuses every customer touchpoint — chat, CRM, support, billing, documents, events — into one governed [context graph](/platform/context-lake/) per subject. - Every fact carries valid-from and valid-to timestamps with provenance back to the source episode — Zep tracks when each fact was true, rather than simply overwriting old information. - On LongMemEval\_S, Zep reports 90.2% accuracy at 104ms retrieval (p50); Supermemory reports 85.2% accuracy, with latency and context size unreported ([results](/research/)). - Governance is enforced in the data layer — entity-level ABAC, retention with legal hold, audit — alongside SOC 2 Type II, HIPAA, and BYOC. The distinction ## Supermemory remembers the conversation. Zep remembers the whole customer. **What Supermemory is.** Supermemory is built for conversational memory — what the user said, across sessions. For a single app that needs to recall its own conversations, that's the appeal. **What Zep is.** Enterprise agents need more. A customer leaves a trail across CRM, support, billing, product events, and documents, and an agent that only remembers chat is working from a fraction of the picture. Zep ingests every source the agent touches and unifies it in one bi-temporal context graph per subject — the [Context Lake](/platform/context-lake/) for AI agents, built on open-source [Graphiti](/platform/graphiti/) and [Konig](/platform/agent-knowledge-graph/), Zep's proprietary graph database service. ### Agent Runtime LangChain · LlamaIndex · CrewAI · Google ADK · custom Any agent framework — or none. The Context Lake is invoked through a single SDK. ### Ingestion chat · JSON · documents · app events Raw signal arrives from any source the agent touches. ### Context Assembly context blocks · templates · token-efficient Relevant context is assembled on demand into token-efficient blocks. ### Graphiti entity extraction · relationships · ontology · invalidation Signal becomes a temporal context graph as new facts arrive and stale ones are invalidated. ### Retrieval sub-200ms · auto-optimized · provenance-linked · policy-filtered Selects what's relevant and what adds the most information within the token budget. ### Governance ABAC · multi-tenant isolation · customer key encryption · retention policies · audit · provenance Native to the data layer, not a layer bolted on. Every read and write is policy-gated for access and provenance; retention runs across the data lifecycle. ### _Konig_ entities · facts & edges · decision traces · episodes Temporal context graph with provenance — sub-200ms retrieval at scale. Benchmarks ## Zep vs. Supermemory on _LongMemEval_ Both systems report results on LongMemEval\_S (500 questions, LLM-as-judge). Zep reports accuracy, retrieval latency, and context size; Supermemory reports accuracy only. Supermemory Zep Scope Conversational memory Every touchpoint — chat, CRM, support, billing, events, documents Data model General-purpose memory store Bi-temporal context graph per subject, with provenance Entities & schema General-purpose store Custom entities and edges, your schema enforced at ingest Governance Account-level Entity-level ABAC, retention with legal hold, audit LongMemEval\_S accuracy 85.2% 90.2% Retrieval latency, p50 Unreported 104 ms Deployment — Managed, BYOK, or BYOC; SOC 2 Type II, HIPAA Scale App-level Millions of context graphs, sub-200ms When to choose ## Pick the layer that fits the agent Stay with Supermemory when Conversational memory for a single app is all you need. - Conversational memory for a single app is all you need - You don't need to fuse business systems — CRM, support, billing — into agent memory - App-level memory and quick integration matter more than entity-level governance and bi-temporal validity Choose Zep when you need Memory that spans every customer touchpoint, governed and served at enterprise scale. - Memory across chat, CRM, support, billing, product events, and documents — not the conversation alone - Bi-temporal facts with point-in-time queries built into the data model - Entity-level governance — ABAC, retention with legal hold, per-request audit - Custom entities and relationships specific to your domain, with your schema enforced - Retrieval that holds at sub-200ms across millions of subjects - Enterprise deployment with SOC 2 Type II, HIPAA, and BYOC Get started ## Ready for agent memory at enterprise _scale_? FAQ ## Frequently asked questions ### What's the difference between Supermemory and Zep? Supermemory is built for conversational memory — what the user said, across sessions. Zep fuses every customer touchpoint — chat, CRM, support, billing, product events, and documents — into one governed, bi-temporal context graph per subject, served in sub-200ms at enterprise scale. ### How do Zep and Supermemory compare on benchmarks? Both report on LongMemEval\_S (500 questions, LLM-as-judge). Zep reports 90.2% accuracy at 104ms retrieval latency (p50); Supermemory reports 85.2% accuracy and does not publish retrieval latency or context size. See the [methodology and results](/research/). ### Can Zep ingest business data beyond chat? Yes. Zep ingests chat, JSON, app events, documents, and business data (CRM, support, billing) through a single SDK and unifies them in one context graph per subject — user, customer, team, or topic. --- ## Zep vs. Vectorize Hindsight: A Neutral Look **Source:** https://www.getzep.com/vectorize-hindsight-alternative/ Zep vs. Vectorize Hindsight Vectorize Hindsight and Zep are both dedicated agent-memory systems that score near the top of LongMemEval. The decision isn't which is more accurate — it's which token-efficiency and operational profile fit your use case. Key takeaways ## The accuracy is roughly equal; the _context cost_ isn't - Both are agent-memory systems that score at the top of LongMemEval (Hindsight reports 91.4%; Zep reports 90.2%) — a single benchmark won't decide your choice. - Hindsight's 91.4% is measured at an **8,192-token retrieval budget**— about 2× the ~4,408 tokens Zep uses for 90.2% on the same benchmark. That extra context is fed to your answer LLM on every call, so at the headline numbers Hindsight costs roughly double the memory tokens per query — real money and added latency at scale. - The other production differences are enterprise governance, managed operations, proven scale, and deployment control, where Zep is purpose-built. (Both systems do temporal reasoning and provenance — see the table.) - Hindsight is MIT open source with a biomimetic memory model that you self-host and operate; its managed/hosted offering is newer. Zep ships a managed [Context Lake](/platform/context-lake/) with SOC 2 Type II, HIPAA, and BYOK/BYOC today. - Pick by need: open-source self-hosting and a human-memory-style model → Hindsight; governed [agent memory](/ai-agents/what-is-agent-memory/) at enterprise scale, at lower per-query context cost → Zep. The distinction ## An open-source self-hosted system vs. a managed Context Lake **What Vectorize Hindsight is.** Hindsight ([vectorize.io](https://vectorize.io/), [GitHub](https://github.com/vectorize-io/hindsight)) is an open-source (MIT) agent-memory system from Vectorize. It organizes memory with biomimetic structures — _World_ (facts), _Experiences_ (the agent's own history), and _Mental Models_(learned understanding formed by reflecting on raw memories) — integrates in about two lines of code, and reports 91.4% on LongMemEval. It offers self-hosting (Docker/embedded), and Vectorize is building a hosted cloud version for managed, production features. **What Zep is.** Zep is a dedicated, managed agent-memory platform — the [Context Lake](/platform/context-lake/). It builds **bi-temporal** context graphs (via the open-source [Graphiti](/platform/graphiti/)) in which every fact carries a validity window and provenance, so the agent reasons over what's true _now_ vs _then_ and can audit any answer to its source. It serves millions of graphs at sub-200ms p95, governs memory in the data layer (ABAC, retention, audit), and deploys managed, BYOK, or BYOC. Zep reports 90.2% on LongMemEval and 94.7% on LoCoMo ([results](/research/)); architecture in the [Zep paper](https://arxiv.org/abs/2501.13956). ### Agent Runtime LangChain · LlamaIndex · CrewAI · Google ADK · custom Any agent framework — or none. The Context Lake is invoked through a single SDK. ### Ingestion chat · JSON · documents · app events Raw signal arrives from any source the agent touches. ### Context Assembly context blocks · templates · token-efficient Relevant context is assembled on demand into token-efficient blocks. ### Graphiti entity extraction · relationships · ontology · invalidation Signal becomes a temporal context graph as new facts arrive and stale ones are invalidated. ### Retrieval sub-200ms · auto-optimized · provenance-linked · policy-filtered Selects what's relevant and what adds the most information within the token budget. ### Governance ABAC · multi-tenant isolation · customer key encryption · retention policies · audit · provenance Native to the data layer, not a layer bolted on. Every read and write is policy-gated for access and provenance; retention runs across the data lifecycle. ### _Konig_ entities · facts & edges · decision traces · episodes Temporal context graph with provenance — sub-200ms retrieval at scale. How they compare ## Vectorize Hindsight vs. Zep, side by side Vectorize Hindsight Zep Model Biomimetic (World / Experiences / Mental Models) Bi-temporal temporal context graph (facts + provenance + validity) LongMemEval (self-reported) 91.4% 90.2% (also 94.7% LoCoMo) Context per query at that score ~8,192 tokens (Budget.HIGH in their runner) ~4,408 tokens — roughly half Temporal reasoning Yes — temporal retrieval arm + temporal indexes on fact lifespans; facts carry temporal links (capped at 20/fact) Bi-temporal edges: “what's true now / what was true then,” automatic fact invalidation, point-in-time queries Provenance Yes — facts trace to the originating message; observations record their source facts Yes — every fact traces to its source episode Open source Yes (MIT), self-hosted Graphiti (the graph library) is open source Managed / hosted Cloud version in development Managed cloud available today Access control No built-in RBAC or ABAC (no users/roles/attribute policies). Static API key, off by default; multi-tenant isolation only via a custom-coded extension. MCP tool allowlisting limits surface area, not per-user access ABAC in the data layer — attribute-based policies govern what each agent/user can read Audit & retention audit\_log table + /audit-logs endpoint, disabled by default; configurable audit retention. No legal hold Audit, retention policies + legal hold in the data layer Compliance / operations Self-hosted OSS — you certify and operate it. No SOC 2 / HIPAA from the vendor Managed service: SOC 2 Type II, HIPAA Deployment Self-host (Docker/embedded) + forthcoming cloud Managed, BYOK, or BYOC (AWS/GCP/Azure) Scale (public) Single Postgres (pgvector/HNSW + BM25 + graph + temporal indexes); stateless API + worker processes scale horizontally; vector search ~10–50ms on 100K+ facts. No published multi-tenant scale figures Millions of graphs per deployment, sub-200ms p95 A note on the benchmark comparison ## Read accuracy alongside tokens, on a matched backbone The published comparison isn't a controlled, matched-backbone head-to-head — by Vectorize's own account. Their repo states that **only Hindsight's score**was independently reproduced (Virginia Tech, The Washington Post) and that “other scores are self-reported by software vendors.” The Zep figure they cite (71.2%) is Zep's _2025-paper_number, not Zep's current 90.2% — so the comparison pairs Hindsight's reproduced Gemini-3 Pro result against Zep's older self-reported figure. Mechanically the two systems are similar: Hindsight extracts facts with an LLM on ingest (“retain”), and on recall runs vector + BM25 + graph + temporal retrieval merged with reciprocal-rank fusion and a cross-encoder reranker, trimmed to a token limit — the same shape as Graphiti/Zep. (Its “LLM-free recall” claim applies to recall, not ingest.) Hindsight publishes a separate speed/cost benchmark but no per-query latency/token figure in the accuracy table. **Context size (from Hindsight's own benchmark code).** Their LongMemEval runner defaults to an **8,192-token retrieval budget at Budget.HIGH** (`thinking_budget=500`), with the answer model at high reasoning effort. That's roughly 1.9× the ~4,408-token context Zep reports on the same benchmark — so the 91.4% is achieved with about double the retrieved context (and a top-tier backbone). On a token-matched basis the gap narrows or reverses. (Hindsight's leaner LoCoMo quality benchmark uses a 4,096-token “low” budget.) Read accuracy alongside latency and tokens, on a matched backbone, against Zep's current numbers. (Hindsight paper: [arXiv 2512.12818](https://arxiv.org/abs/2512.12818).) When to choose ## Pick the tool that fits the problem Choose Vectorize Hindsight when You want an open-source system you can self-host or embed, you like the biomimetic World / Experiences / Mental-Models model, and your priority is LongMemEval-style recall. - Open-source self-hosting or embedding is a hard requirement - You prefer the biomimetic, human-memory-style model - LongMemEval-style recall is the top priority, and the MIT license makes it easy to try Choose Zep when Memory has to be governed and operated at enterprise scale today, at lower per-query context cost. - Bi-temporal reasoning and provenance for auditability - Attribute-based access control and retention in the data layer - SOC 2 Type II / HIPAA and deployment control (managed, BYOK, or BYOC) - Proven performance across millions of graphs at sub-200ms p95 - S&P Global Market Intelligence (451 Research) initiated coverage on Zep as a likely de facto partner in the enterprise agent stack Get started ## Add governed agent memory at _half the context cost_ FAQ ## Frequently asked questions ### Is Hindsight or Zep more accurate? Both score near the top of LongMemEval (Hindsight 91.4%, Zep 90.2%; Zep also reports 94.7% on LoCoMo) — effectively a tie. The more useful question is _at what cost_: Hindsight's number is measured at an ~8,192-token retrieval budget vs Zep's ~4,408 — about double the context your answer model processes on every query. For the same accuracy, that's roughly 2× the memory-token cost and added latency at scale. ### Which is cheaper to run per query? At the published accuracy levels, Zep feeds your answer LLM about half the memory tokens Hindsight does (~4,408 vs ~8,192), so per-query token cost and latency are lower. Hindsight's recall path itself is LLM-free and fast (100–600ms); the cost difference is in how much retrieved context each system hands to the answer model. ### Is Hindsight open source? Yes, MIT-licensed. Zep's graph library, Graphiti, is also open source. Zep's managed platform, including Konig, its proprietary graph database service, is commercial. ### Which is better for enterprise? Zep is purpose-built for governed memory at scale (ABAC, retention, audit, SOC 2 Type II, HIPAA, BYOK/BYOC, millions of graphs). Evaluate both against your governance, deployment, and scale requirements. ### Does Hindsight support RBAC, ABAC, and audit logging? Hindsight has no built-in RBAC or ABAC — no users, roles, or attribute-based access policies. Its built-in auth is a single static API key (off by default); multi-tenant isolation requires coding a custom extension, and the only finer-grained controls are MCP tool allowlisting (which tools are exposed) and a config-field permission hook. Audit logging exists but is disabled by default, and there's no legal hold or vendor compliance certification. Zep provides ABAC, retention with legal hold, and audit in the data layer, as a managed SOC 2 Type II / HIPAA service. If access control and auditability are requirements, that's a meaningful gap to weigh. ### Can I self-host? Hindsight supports self-hosting today. Zep offers managed cloud, BYOK, and BYOC (in your VPC); Graphiti can also be self-hosted standalone. --- ## Zep vs. Vertex AI Memory Bank: Neutral Alternative **Source:** https://www.getzep.com/vertex-ai-memory-bank-alternative/ Zep vs. Vertex AI Memory Bank Vertex AI Memory Bank is Google's managed memory service for agents in Vertex AI Agent Engine. Zep is a neutral, multi-LLM, multi-cloud Context Lake that manages, governs, and serves agent memory on temporal context graphs. Key takeaways ## A hyperscaler primitive, or a _neutral_ memory layer - Vertex AI Memory Bank ([docs](https://cloud.google.com/vertex-ai/generative-ai/docs/agent-engine/memory-bank/overview)) is bound to Google Cloud/Vertex; Zep is a neutral, multi-LLM, multi-cloud [Context Lake](/platform/context-lake/). - The decision is lock-in vs. neutrality: keep one memory strategy across Gemini, OpenAI, Anthropic, and Meta. - Zep runs managed, BYOK, or BYOC on AWS/GCP/Azure, builds bi-temporal context graphs, and reports 94.7% LoCoMo and 90.2% LongMemEval ([results](https://www.getzep.com/research/)). The distinction ## A Google Cloud primitive vs. a neutral layer **What Vertex AI Memory Bank is.** Memory Bank is part of Google's Vertex AI Agent Engine. It extracts and stores memories for agents built on Vertex, and surfaces them back at run time, integrated with the Google Cloud agent stack. For teams standardized on Vertex/GCP, the native integration is the draw. **What Zep is.** Zep is a dedicated, neutral memory layer — the [Context Lake](/platform/context-lake/) for AI agents. It builds bi-temporal context graphs from chat, business data, and documents (open-source [Graphiti](/platform/graphiti/) on Konig, Zep's proprietary graph database service), serves token-efficient context in sub-200ms p95, and deploys as managed cloud, BYOK, or BYOC on the cloud you choose. It's model- and framework-agnostic. ### Agent Runtime LangChain · LlamaIndex · CrewAI · Google ADK · custom Any agent framework — or none. The Context Lake is invoked through a single SDK. ### Ingestion chat · JSON · documents · app events Raw signal arrives from any source the agent touches. ### Context Assembly context blocks · templates · token-efficient Relevant context is assembled on demand into token-efficient blocks. ### Graphiti entity extraction · relationships · ontology · invalidation Signal becomes a temporal context graph as new facts arrive and stale ones are invalidated. ### Retrieval sub-200ms · auto-optimized · provenance-linked · policy-filtered Selects what's relevant and what adds the most information within the token budget. ### Governance ABAC · multi-tenant isolation · customer key encryption · retention policies · audit · provenance Native to the data layer, not a layer bolted on. Every read and write is policy-gated for access and provenance; retention runs across the data lifecycle. ### _Konig_ entities · facts & edges · decision traces · episodes Temporal context graph with provenance — sub-200ms retrieval at scale. How they compare ## Memory Bank vs. Zep, side by side Vertex AI Memory Bank Zep Ecosystem Bound to Google Cloud / Vertex Neutral — any model, any cloud Model providers Google-centric (Gemini) OpenAI, Anthropic, Meta, Gemini, others Memory model Memories extracted from session history (Agent Engine Sessions + Memory Bank) Bi-temporal context graph (provenance + validity) Temporal reasoning No — extraction-based; no temporal graph “What's true now / what was true then,” auto fact invalidation Deployment GCP / Vertex Managed, BYOK, or BYOC (AWS/GCP/Azure) Benchmarks — 94.7% LoCoMo (155ms), 90.2% LongMemEval (162ms) Lock-in risk Higher (ecosystem-bound) Lower (portable across stacks) The strategic question ## Lock-in vs. _neutrality_ The same logic S&P Global Market Intelligence flagged for hyperscaler primitives applies here: your agents' memory is among the most valuable and sticky data you own. A bundled memory service keeps it inside one cloud's ecosystem. A neutral layer lets you keep one consistent context strategy across Gemini, OpenAI, Anthropic, and Meta — and move between clouds without re-platforming your memory. When to choose ## Pick the tool that fits the strategy Choose Memory Bank when You're standardized on Google Cloud and Vertex, use Gemini as your primary model, and the bundled memory meets your needs. - Standardized on Google Cloud and Vertex - Gemini is your primary model - Bundled memory meets your needs Choose Zep when You want neutrality across models and clouds, with governed memory at scale. - Neutrality across models and clouds - Temporal, provenance-tracked, governed memory (ABAC, retention, audit) - Regulated workloads with BYOK/BYOC deployment control - Benchmark-proven retrieval quality at enterprise scale Get started ## Keep your agent memory _portable_ FAQ ## Frequently asked questions ### Is Vertex AI Memory Bank enough for enterprise agent memory? If you're committed to Vertex/GCP and Gemini and need managed memory, it can be. For neutrality, temporal reasoning, provenance, and portable governance, evaluate a dedicated layer like Zep. ### Can Zep run on Google Cloud? Yes — managed, with your own keys, or inside your own VPC on GCP (or AWS/Azure). ### Does Zep work with Gemini? Zep is model-agnostic and works with Gemini as well as OpenAI, Anthropic, and others — so memory isn't tied to one model provider. ---