Zep Memory is live · free plan, paid from $125/mo · last verified 26 Oct 2026 · viability healthy
Zep Memory
Build governed AI agent memory with Zep's temporal context graph and dated, traceable facts
Pick Zep when agent context has to hold up in a compliance review: dated fact invalidation, provenance back to the source Episode, and ABAC enforced in the data layer are engineering you don't bolt onto a SQLite memory file. The v4 Context API and hyperedges shipped in October 2026 make it a stronger fit for multi-entity reasoning than earlier versions. Budget carefully — credits meter Episode ingestion, so a chatty agent is an expensive agent, and Observations, webhooks, and analytics only
Verdict generatedVerified 4d ago · liveness 81/100 · cite: rightaichoice.com/tools/zep-memory
- Enterprises running production agents where memory must survive an audit
- Regulated teams needing SOC 2 Type II, HIPAA BAA, and EU DPA coverage
- Teams that must deploy in their own VPC or encrypt with their own KMS keys
- Agent platforms serving many end users, accounts, or business domains at once
- Simple RAG or Q&A over a static document set
- Stateless agents that carry no context between sessions
- High-chat-volume, low-value-per-message bots that burn credits on Episode ingestion
We scan live Reddit threads, YouTube comments, X posts, G2 reviews and other communities — and hand you an honest verdict in under a minute.
- Honest verdict, not marketing
- Real pros & cons from real users
- Attributed quotes with receipts
3 free scans · no card needed
Skip Zep if you need conversational recall over a static document set and nothing else — the provenance, retention, and access-policy machinery only pays off when agent context has to be governed and defensible.
Credits are consumed per Episode by size — 1 credit up to 350 bytes and another per extra 350 bytes — so long JSON payloads and verbose chat turns cost more than the credit count suggests.
Free at 10,000 credits/mo suits prototypes; Flex at $125/mo billed monthly fits a funded startup with a few production agents, and Flex Plus at $375/mo billed monthly is where most real feature needs land because Observations, webhooks, and analytics live there. Annual billing saves 17%. Expect Enterprise to exceed that, since guaranteed rate limits, 1-year audit logs, and BYOC are only negotiated. For plain conversational recall, lighter memory libraries cost less.
In short
Zep Memory — Build governed AI agent memory with Zep's temporal context graph and dated, traceable facts. Best for Enterprises running production agents where memory must survive an audit, Regulated teams needing SOC 2 Type II, HIPAA BAA, and EU DPA coverage, Teams that must deploy in their own VPC or encrypt with their own KMS keys. Free to start; paid plans from $125/mo.
No head-to-head comparison for Zep Memory yet — the alternatives below are the closest matches we list.
What is Zep Memory?
Zep Memory — Build governed AI agent memory with Zep's temporal context graph and dated, traceable facts. Best for Enterprises running production agents where memory must survive an audit, Regulated teams needing SOC 2 Type II, HIPAA BAA, and EU DPA coverage, Teams that must deploy in their own VPC or encrypt with their own KMS keys. Free to start; paid plans from $125/mo.
Zep is a context layer for AI agents that unifies business data, documents, and conversations into governed Context Graphs. Each graph is built per user, customer, account, or domain, and retrieval returns only the facts a task depends on rather than pulling whole records into the prompt. Zep puts that at roughly 90% less token use than tool calling. Facts are dated and update as source data changes; when new information contradicts an old fact, the old one is invalidated and kept as queryable history, and each fact traces back to the Episode that produced it. Governance lives in the data layer, not in glue code. Attribute-based access control policies attach to API keys, retention schedules and legal hold control deletion, and audit logs record every request and access decision. The Context Lake unifies millions of Context Graphs on Konig, Zep's own graph database service, with p95 retrieval stated at 148ms at 10K graphs through 168ms at 100M — under 200ms regardless of size. As of October 2026 the platform runs on the v4 Context API, which lets agents traverse the graph by connections and inspect how context was assembled, with stable error codes; v3 remains supported during migration. Hyperedges keep facts about three or more entities intact as one unit so retrieval doesn't return a partial pair. A Context MCP Server gives each user one governed memory shared across Claude, ChatGPT, and in-house agents, and deployment options span managed cloud, BYOK, or your own VPC. Pricing is credit-based — 1 credit per Episode up to 350 bytes, with retrieval, storage, threads, and users unmetered. Teams running agents where context has to survive an audit are the core audience; lighter memory libraries cover simpler cases at less cost.
Key features
- Temporal context graphs with dated facts updated as source data changes
- Automatic invalidation of contradicted facts, retained as queryable history
- Point-in-time queries asking what was true on any past date
- Provenance tracking from each derived fact back to its source Episode
- Context assembly from business data, documents, and conversations
- Context Lake unifying millions of Context Graphs as one managed system
- Konig graph database service powering the Zep data plane
- Hyperedges storing facts about three or more entities as one intact unit
- v4 Context API with graph traversal by connections and assembly inspection
- Stable error codes in the v4 Context API (v3 supported during migration)
- Attribute-based access control with policies attached to API keys
- Retention schedules, data expiry, and legal hold inside the data layer
- Audit and API logs of every request and access decision
- Context MCP Server sharing one governed memory across Claude, ChatGPT, and in-house agents
- Plugin for Claude Code, Codex, and Cursor bundling the Zep docs MCP server
Behind the verdict
The question with any agent memory layer is whether it survives contact with production. Zep's answer is governance baked into the data plane: access policies, retention, legal hold, and audit logs sit alongside the facts rather than in application code. If your agents touch customer records or regulated data, that separation is the whole argument. In practice we'd reach for Zep when an auditor asks who saw what and when — provenance from each fact back to its source Episode makes that answerable. The temporal model is the other draw. Facts carry dates, contradictions invalidate the old fact but keep it as history, and point-in-time queries let you ask what was true on a past date. That matters for renewals, account context, and anything where state changes over months. The October 2026 v4 Context API builds on it with graph traversal by connections and assembly inspection, and hyperedges prevent the partial-fact problem where a relationship involving three parties gets split into pairs. Where it bites: credits meter Episode ingestion, not retrieval. A high-chat-volume bot that fires thousands of small messages daily will burn through the 50,000 included credits on the $125/mo Flex plan fast, and the overage rate is $25 per 10,000 credits. The features most teams want next — Observations, custom extraction instructions, webhooks, analytics — only arrive on Flex Plus at $375/mo. Free gets you 10,000 credits to prototype, but no rollover and tight project limits. Compared with lighter libraries and memory APIs, you're paying for the governance stack and the Context Lake. Konig keeps retrieval under 200ms at 100M graphs, which is the scale claim that separates it from a self-hosted vector store plus a cron job. If your agents are internal prototypes or stateless Q&A
Models under the hood
as of 2026-10-23
Limitations
- Self-serve plans are credit-metered: 1 credit per Episode up to 350 bytes, plus 1 credit per additional 350 bytes or part, and 1/8 credit per webhook invocation; retrieval, storage, threads, users, and graph storage are unmetered.
- Flex is $125/mo billed monthly (annual saves 17%) with 50,000 credits, 5 projects, 5 Context MCP Server seats, 10 custom entity and edge types, and 1 day of API logs.
- Flex Plus is $375/mo billed monthly with 200,000 credits, 10 projects, 15 MCP seats, 20 custom types, and 7 days of API logs.
- The Free tier's 10,000 monthly credits do not roll over or auto top-up, and Free and Flex rate limits may be lowered depending on service usage.
- Observations, webhooks, analytics, and custom extraction instructions are Flex Plus and above.
- Guaranteed rate limits with SLA, unlimited projects, 1-year audit and API logs, SOC 2 Type II, HIPAA BAA, EU DPA, and BYOK or BYOC deployment are Enterprise-only and custom-quoted.
as of 2026-10-21
Researching Zep Memory? Get your full AI stack in 60 seconds.
Free, no signup — tell us your goal and get tools matched to your budget & existing stack.
Is Zep Memory still active?
Zep Memory is live · free plan, paid from $125/mo · last verified 26 Oct 2026 · viability healthy.
Viability score
How well maintained and how widely used is Zep Memory? Built from what the vendor actually publishes (docs, changelog, tutorials, integrations, pricing), whether the site is live, and how much real users discuss it. How we calculate this
Last calculated: October 2026
How we score →What's new in Zep Memory
Checked 9 days agoAcross the latest 5 updates: 2 feature updates, 1 launch and 2 news mentions.
Everything we shipped in the Zep Context API
Zep rounds up every feature in the v4 Context API and explains how each helps agents make decisions from Context Graphs.
Hyperedges keep multi-entity facts intact in Zep context graphs
Zep hyperedges store facts about three or more entities as one unit, so retrieval does not return a partial pair and miss the rest.
Zep v4 ships as the Context API for agents
Zep v4 lets agents traverse the graph by connections and inspect how it was built; errors carry stable codes. v3 stays supported during migration.
Zep publishes guidance on defending agent memory against poisoning
Post covers how persistent memory makes prompt injection durable and which application and Zep controls limit memory poisoning.
Zep explains Konig, its graph database service for agent memory
Zep details Konig, the graph database data plane under its memory platform, and the engineering lessons behind it.
What do users say about Zep Memory?
What people actually say about Zep Memory — is it worth it?
We ran a structured research pass across product reviews, community discussions, and post-purchase forum threads to surface the patterns vendors won't publish themselves. Below: the recurring strengths, the hidden costs people mention most, and the cohort that consistently regrets adopting this tool.
ResearchedDiscussed by users across 3 sources (Hacker News, YouTube, Lemmy) · researched Aug 14, 2026.
Average across the 3 sources that answered — each source counts once, not each post.
- +Temporal context graphs handle contradictions and track fact evolution.
- +Sub-200ms retrieval at scale supports production agents with low latency.
- +Provenance preservation traces each fact to its source episode.
- +Built-in governance: ABAC, retention policies, legal hold, audit logs.
- +Integrates with Python, TypeScript, and Go — flexible for any stack.
- −LLM calls on every turn drive token costs and third-party exposure.
- −Steep learning curve; documentation is sparse for advanced setup.
- −Little community troubleshooting; user questions often unanswered.
- −Overkill for simple agent memory; lighter tools are cheaper.
- −Pricing transparency is weak; hidden costs for Flex Plus tiers.
- • Tokens for LLM fact extraction on every turn
- • Potential overage charges for high volume API calls
- • Enterprise support likely requires a premium contract
How we verified this
Every listed fact about Zep Memory— pricing, features, links, status — is re-read from the vendor's own pages by our verification pipeline and each change is logged with a date; the last pass ran on 26 Oct 2026. Pages mix model-assisted synthesis with human review; see how we verify.
Verification history
We have re-verified Zep Memory 17 times since . Each pass re-reads the vendor's own pages and re-checks every listed field against that evidence; passes where nothing had changed are marked as such.
- — changed SSL certificate
- — changed description, who should skip it, features, summary, pricing tiers, our verdict
- — changed pricing page
- — changed summary, setup time, pricing fit, features, hidden costs, who it suits, pricing tiers, description, integrations, limitations, switching in, who should skip it, switching out, use cases, our verdict
- — changed limitations, models, pricing page, platforms
- — changed pricing tiers, who it suits, description, our verdict, features, integrations, who should skip it, summary
Showing the 6 most recent of 17 verification passes.
Free to cite with attribution — this page re-verifies continuously.
How much does Zep Memory cost in 2026?
Zep Memory: Free to start; paid plans from $125/mo. Pricing last verified 26 Oct 2026.
Plans compared
Every published Zep Memory tier, its price as the vendor publishes it, who it fits, and what it adds over the tier below. The 12-month cost projection follows.
| Plan | Price | Billing | Who it's for |
|---|---|---|---|
| Free | Free | — | Solo developer or small team prototyping one agent and testing whether Context Graphs fit the use case.Adds: Starting tier: 10,000 credits per month, 2 projects, 1 Context MCP Server seat, unlimited memories and retrieval, no rollover or auto top-up. |
| Flex | $125/moVendor: “$125/mo billed monthly” | monthly | Funded startup running a handful of production agents that need predictable throughput and credit rollover.Adds: Adds 50,000 credits/mo (then $25 per 10,000), 20% auto top-up, 30-day rollover, 600 requests/min, 5 projects, 5 MCP seats, 10 custom entity and edge types. |
| Flex Plus | $375/moVendor: “$375/mo billed monthly” | monthly | Team that needs pattern detection, event webhooks, and analytics across more projects and higher request throughput.Adds: Adds Observations, webhooks, analytics, custom extraction instructions, 200,000 credits/mo (then $75 per 40,000), 1,000 requests/min, 10 projects, 15 MCP seats, 7-day API logs. |
| Enterprise | Custom | quote | Regulated enterprise deploying inside its own VPC or with customer-managed keys and a guaranteed SLA.Adds: Adds SOC 2 Type II, HIPAA BAA, EU DPA, guaranteed rate limits with SLA, unlimited projects, 1-year audit and API logs, and Cloud, BYOK, or BYOC deployment. |
12-month cost
Project the real annual outlay, including the implied monthly cost when only an annual tier is published.
Vendor list price only. Add-on usage, seat overages, and contract minimums are surfaced under Hidden costs & gotchas.
Where the pricing makes sense
The company stage and team size where Zep Memory's pricing actually pencils out — and where peers do it cheaper.
Free at 10,000 credits/mo suits prototypes; Flex at $125/mo billed monthly fits a funded startup with a few production agents, and Flex Plus at $375/mo billed monthly is where most real feature needs land because Observations, webhooks, and analytics live there. Annual billing saves 17%. Expect Enterprise to exceed that, since guaranteed rate limits, 1-year audit logs, and BYOC are only negotiated. For plain conversational recall, lighter memory libraries cost less.
Has Zep Memory's pricing changed?
No plan price for Zep Memory has moved since we began logging in Aug 2026; the pricing page is re-read on every verification pass.
Who should pick Zep Memory?
Pick Zep Memory if you are Enterprises running production agents where memory must survive an audit, Regulated teams needing SOC 2 Type II, HIPAA BAA, and EU DPA coverage, Teams that must deploy in their own VPC or encrypt with their own KMS keys, Agent platforms serving many end users, accounts, or business domains at once.
Look elsewhere if you are Simple RAG or Q&A over a static document set, Stateless agents that carry no context between sessions, High-chat-volume, low-value-per-message bots that burn credits on Episode ingestion, Prototypes where governance, audit trails, and provenance add no value.
Skip it if Skip Zep if you need conversational recall over a static document set and nothing else — the provenance, retention, and access-policy machinery only pays off when agent context has to be governed and defensible.
GeneratedReal-world workflow fit
Concrete scenarios for the personas Zep Memory actually fits — and what changes day-one when you adopt it.
You wire the v4 Context API into a support agent, sending each ticket message as an Episode and letting Zep build a Context Graph per customer.
Outcome: The agent answers from current, dated facts and you get a 50,000-credit Flex allowance at $125/mo billed monthly for the first production month.
You deploy Zep in your own VPC via BYOC, attach ABAC policies to per-team API keys, and set a 30-day retention window with legal hold on regulated accounts.
Outcome: Audit and API logs kept for 1 year give you a defensible trail without building a separate governance service.
You stand up the Context MCP Server so a Claude assistant and an in-house agent read the same governed memory for each end user.
Outcome: One customer-facing memory shared across tools, which on Flex Plus at $375/mo billed monthly covers 15 MCP seats and 1,000 requests per minute.
Use cases
- Enterprise agent answering account questions from current facts and tracing each answer to its source Episode.
- Support agent that remembers a customer's past issues, preferences, and account status across sessions.
- Deal-flow or portfolio agent reasoning over months of dated context with point-in-time accuracy.
- Voice agent retrieving relevant business data at under 200ms p95 during live support.
- Regulated deployment needing ABAC, retention policy, legal hold, and audit logs inside the data layer.
- Multi-agent platform where every agent shares one governed context graph per customer or account.
- Personal shopping assistant tracking brand preferences, purchase history, and warranty status.
Setup time & first value
How long it actually takes to get something useful out of Zep Memory — broken out by persona, not the marketing-page minute.
A developer can send a first Episode and see a populated Context Graph within an afternoon on the Free tier. Getting to production — ABAC policies, retention schedules, and BYOC deployment — is a multi-week project for a platform team, mostly spent on policy design rather than integration.
Switching to or from Zep Memory
How to bring data in from common predecessors and how to get it back out — written for the switcher, not the buyer.
- →From a vector store memory layer: send existing conversation logs as Episodes so Zep derives dated facts instead of embedding raw turns.
- →From v3 API: keep v3 running during the move to the v4 Context API, which adds graph traversal by connections and stable error codes.
- →From Zep Free: upgrade to Flex at $125/mo billed monthly when the 10,000 credits or 2-project cap becomes the constraint.
- →From Flex to Flex Plus: move up at $375/mo billed monthly to unlock Observations, webhooks, analytics, custom extraction instructions, and 7-day API logs.
- ↗To a lighter memory library: export graph facts and Episodes, then re-derive embeddings in the new store.
- ↗To a self-hosted graph stack: use BYOC as the exit path if the constraint is the managed control plane rather than Zep's data model.
- ↗To Enterprise on Zep: negotiate custom credits and guaranteed rate limits instead of migrating to a different vendor.
Integrations
Resources & guides
Tutorials & Learning
YouTube returned 6 videos for “Zep Memory”, and we withheld 6: 6 did not mention Zep Memory. We are showing none, because we could not prove any of them are about Zep Memory.
Official links
What are the best Zep Memory alternatives?
The closest alternatives to Zep Memory we list are Cognee, Mem0 and Vectorize. Ranked by direct product-type match, not category overlap.
| Alternative | Pricing | Starts at | Last verified |
|---|---|---|---|
| Cognee | Freemium | Pricing not published | — |
| Mem0 | Freemium | Pricing not published | — |
| Vectorize | Freemium | Pricing not published | — |
Alternatives to Zep Memory
View all
Tools that pair well with Zep Memory
Common stack mates teams adopt alongside Zep Memory, with the specific reason each pairing earns its keep.
Cognee
Cognee gives AI agents persistent knowledge-graph memory with cited recall across sessions, teams, and tools.
Mem0
Mem0 is a memory layer that gives AI agents and apps persistent context across sessions and users
Vectorize
Agent memory that learns from failed tool calls and user corrections, not just stores facts
Frequently asked questions
Categories
Best-of guides
Topics
Used Zep Memory? Help shape our editorial sentiment research.