Visit website
  • @zep_ai
  • metaskills
  • @getzep
  • @zepai_videos

Zep stores agent memory in temporal context graphs built from chat history, business data and user interactions, invalidating superseded facts and returning assembled context with p95 retrieval under 200ms. Engineering teams shipping production agents call Zep from Python, TypeScript and Go SDKs, with SOC 2 Type II and BYOC deployment.

Zep screenshot 1 of 2

1 / 2

Key facts about Zep

Verified facts from the vendor site · Last verified Aug 22, 2026. Independent Zep review covering pricing, features, who it is for, and alternatives.

Product
Zep
Company
Zep AI, Inc.
Pricing
Zep starts at $0/month.
Free tier
Zep has a free tier.
Founded
2023
Platforms
Web
Suitable for
Teams giving a production customer-support agent persistent memory of a user's prior issues and stated preferences and fusing CRM, billing and product-event JSON into a single per-user graph an agent can query

Quick answers about Zep

What is Zep?

Zep Software, Inc. describes Zep as a "Context Lake" — the data-lake pattern applied to agent context, where millions of individual context graphs are governed and served as one system.

How much does Zep cost?

Zep is freemium. It starts at $0/month. Paid plans include Free: monthly $0/mo: yearly $0/yr ($0/mo effective); Flex: monthly $125/mo: yearly $1250/yr ($104.17/mo effective, save 17%); Flex Plus: monthly $375/mo: yearly $3750/yr ($312.5/mo effective, save 17%). Check getzep.com for current prices.

Is Zep free?

Zep has a free tier. Paid plans start at $125/month.

Who is Zep for?

Zep is best for teams giving a production customer-support agent persistent memory of a user's prior issues and stated preferences and fusing CRM, billing and product-event JSON into a single per-user graph an agent can query.

Should I use Zep?

Zep is best for teams giving a production customer-support agent persistent memory of a user's prior issues and stated preferences and fusing CRM, billing and product-event JSON into a single per-user graph an agent can query. Skip it if zep plan limits come from the published pricing table..

What are the main Zep features?

The main Zep features are Temporal context graphs that record facts with a validity window, Fact invalidation when new information contradicts the graph, with old facts kept as history, Point-in-time queries: ask what is true now or what was true on a past date, Multi-source ingest from chat history, business data (JSON) and user interactions, and Automated context assembly returning token-efficient context blocks.

What are the best Zep alternatives?

The closest Zep alternatives on Citeware are Pollinations.AI, Pinokio, Blackbox, Seldon, Snorkel AI. Full list: https://citeware.io/alternatives/zep.

What are the downsides of Zep?

Limitations called out on this listing: Zep plan limits come from the published pricing table.; Zep packaging is recorded from the vendor pages available at analysis time..

Does Zep work on web and desktop?

Zep is available on Web.

Zep alternatives and comparisons

Recent discussion

Public threads about Zep

Community observations

No recent Hacker News threads

Nothing matched Zep on Hacker News right now. Search X, Reddit, or another network above.

About Zep

Citeware analysis

CategoryAI Infrastructure & APIs

PricingStarts at $0/month

Free tierYes

Zep Software, Inc. describes Zep as a "Context Lake" — the data-lake pattern applied to agent context, where millions of individual context graphs are governed and served as one system. Rather than replaying a growing transcript into every model call, Zep ingests each Episode (a chat message, a JSON payload, or a block of text), extracts entities and facts into a graph, and hands back a compact context block on request. The Zep homepage publishes p95 retrieval latency of 148ms at 10K graphs rising only to 168ms at 100M, and states that retrieval with Zep "stays under 200 milliseconds, regardless of graph size or count."

Temporality is the organising idea in Zep. When incoming information contradicts an existing fact, Zep invalidates the old fact instead of deleting it, so an application can ask Zep both what is true now and what was true on a given past date. Zep also surfaces what the company calls Observations — patterns, recurrences and co-occurrences detected in the shape of the graph, such as a customer who upgrades within two weeks of each product launch. Every fact inside a Zep graph traces back to the source episode that produced it, which is what makes an answer auditable rather than merely plausible.

Day-to-day work with Zep is deliberately small in surface area: add messages to a thread with return_context enabled, push structured business data into a user's graph with graph.add, and pull assembled context with get_user_context. Zep ships SDKs for Python, TypeScript and Go, and the vendor states Zep works with any agent framework or none at all. Around those calls, Zep adds a Memory MCP Server with seats metered per plan, webhooks, analytics, and a dashboard reporting episodes added, graphs and users created, average latency and error rate.

Zep bills on credits rather than seats or stored bytes. One credit covers an Episode up to 350 bytes, with one further credit for each additional 350 bytes or part thereof, while webhook invocations cost 1/8 credit where available and retrieval, storage, threads, users and graph storage are billed at zero credits. The self-serve Flex plan on the Zep pricing page is $1,250 per year, shown as $104 per month billed annually and as $125 per month on the plan comparison table, and includes 50,000 credits a month with overage at $25 per 10,000 credits. Flex Plus is $3,750 per year, shown as $312 per month billed annually and $375 per month on the comparison table, with 200,000 credits a month and $75 per 40,000 credits of overage.

Enterprise pricing at Zep is quoted rather than published: custom credit volumes at negotiated rates, guaranteed rate limits with an SLA, unlimited projects, custom Memory MCP Server seats, one-year API log retention, and Slack or Teams support with a dedicated account manager. Zep additionally advertises an "emerging price" for fast-growing venture-capital funded startups. Deployment scales alongside the plan — managed Zep Cloud on every tier, Cloud plus customer-held KMS keys on AWS, GCP or Azure as BYOK, and Bring Your Own Cloud where the Zep service runs inside the customer's own VPC.

The constraints in Zep deserve a close read before commitment. Zep credits roll over for 30 days on Flex and 60 days on Flex Plus, while free-plan credits do not roll over at all; the Zep FAQ also warns that Free and Flex accounts may see rate limits lowered depending on service-wide load, with lower-priority Episode processing on the free tier. Observations, webhooks, analytics and custom extraction instructions are withheld from Flex and appear only from Flex Plus upward, and Flex retains API logs for a single day against seven days on Flex Plus.

Buyers should also note a documented inconsistency on the Zep pricing page itself: the plan comparison table places SOC 2 Type II, HIPAA BAA, audit logs and the EU DPA in the Enterprise column with dashes against Flex and Flex Plus, while the Zep FAQ directly below answers that Zep is SOC 2 Type II certified and that Zep signs Data Processing Agreements with EU customers. Both statements come from Zep and both are recorded here; a compliance reviewer on a self-serve plan should get the applicable scope in writing.

Against peers in agent memory and retrieval infrastructure, Zep leans on structure and governance rather than on being the cheapest key-value memory store. Zep is built on the open-source Graphiti framework and a proprietary Context Graph Engine, publishes LoCoMo (94.7% accuracy, 155ms retrieval, 5,760 tokens) and LongMemEval (90.2% accuracy, 162ms retrieval, 4,408 tokens) benchmark figures, and maintains a dedicated mem0-alternative page. Teams that only need a scratchpad between turns will find Zep heavier than necessary; teams that must explain where an agent's belief came from, and when it changed, are the audience Zep is written for.

Suitable for

Teams giving a production customer-support agent persistent memory of a user's prior issues and stated preferences and fusing CRM, billing and product-event JSON into a single per-user graph an agent can query

  • Giving a production customer-support agent persistent memory of a user's prior issues and stated preferences
  • Fusing CRM, billing and product-event JSON into a single per-user graph an agent can query
  • Answering point-in-time questions such as what a customer's plan or preference was on a specific date
  • Reducing prompt token spend by replacing full chat transcripts with an assembled context block
  • Auditing why an agent said something by tracing the underlying fact back to its source episode
  • Detecting behavioural patterns, such as repeat upgrade timing, and feeding them to an agent as Observations

Evidence: Homepage (Primary) · Pricing (Primary) · About (Primary) · Verified Aug 22, 2026.

Pros and cons

Citeware analysis

Pros

  • Zep publishes concrete p95 latency figures across graph sizes (148ms at 10K to 168ms at 100M) rather than a vague speed claim
  • Z
  • Temporal context graphs that record facts with a validity window
  • Fact invalidation when new information contradicts the graph, with old facts kept as history
  • Point-in-time queries: ask what is true now or what was true on a past date
  • Multi-source ingest from chat history, business data (JSON) and user interactions
  • Automated context assembly returning token-efficient context blocks
  • Observations: patterns, recurrences and co-occurrences detected across the graph

Cons

  • Zep plan limits come from the published pricing table.
  • Zep packaging is recorded from the vendor pages available at analysis time.

Evidence: Homepage (Primary) · Verified Aug 22, 2026.

Features

Verified facts

  • Temporal context graphs that record facts with a validity window
  • Fact invalidation when new information contradicts the graph, with old facts kept as history
  • Point-in-time queries: ask what is true now or what was true on a past date
  • Multi-source ingest from chat history, business data (JSON) and user interactions
  • Automated context assembly returning token-efficient context blocks
  • Observations: patterns, recurrences and co-occurrences detected across the graph
  • Provenance on every fact, traced back to the source episode
  • Context Lake architecture governing millions of context graphs as one system
  • Proprietary Context Graph Engine runtime over a three-tier data layer
  • Sub-200ms p95 retrieval published from 10K (148ms) to 100M (168ms) graphs
  • Attribute-based access control across agents, users and context sources
  • Policy-driven retention with legal hold that blocks deletion

Evidence: Homepage (Primary) · Verified Aug 22, 2026.

Pricing

Verified facts

Free

$0/mo

  • 10,000 credits per month (no rollover or auto top-up)
  • 2 projects
  • Memory MCP Server seat
  • Custom entity/edge types
  • Variable rate limits depending on service-wide load
  • Lower priority Episode processing
  • Community support

Flex

$125/mo

Billed monthly

  • 50,000 credits per month included, then $25 per 10,000 credits
  • Auto top-up at 20% (10,000 credits / $25)
  • 30-day credit rollover
  • 600 requests per minute
  • 5 projects
  • 5 Memory MCP Server seats
  • 10 custom entity/edge types
  • API logs retained 1 day
  • Unlimited memories and retrieval users
  • Community support
  • Cloud deployment

Flex Plus

$375/mo

Billed monthly

  • 200,000 credits per month included, then $75 per 40,000 credits
  • Auto top-up at 20% (40,000 credits / $75)
  • 60-day credit rollover
  • 1,000 requests per minute
  • 10 projects
  • 15 Memory MCP Server seats
  • 20 custom entity/edge types
  • Observations, custom extraction instructions, webhooks, analytics
  • API logs retained 7 days
  • Unlimited memories and retrieval users
  • Priority support

Evidence: Homepage (Primary) · Pricing (Primary) · Verified Aug 22, 2026.

Policies

Verified facts

The company behind Zep is Zep AI, Inc.

Operates a cloud-based long-term memory and context-engineering software platform for AI agents and assistants (the "Platform"). Zep Service includes APIs, software development kits, and a web-based dashboard, provisioned with API Key(s) for programmatic access. Enterprise Fees are non-refundable except as expressly set out in the Agreement or an Order Form, and are not eligible for set off. Zep also lists other losses and interruptions it does not cover.

Zep says it collects account and identity, contact details, product usage and analytics, device and ip data, and cookies and tracking. That data is used for registering with the Service by entering email address, user name and password.

No dedicated refund page was found. The Terms of Service state that for Enterprise tier subscriptions all Fees are non-refundable, except as expressly set out in the Agreement or an applicable Order Form, and are not eligible for set off. Self-serve subscription Fees are prepaid at the start of each billing period; no refund entitlement is stated.

Operates a cloud-based long-term memory and context-engineering software platform for AI agents and assistants (the "Platform"). Zep Service includes APIs, software development kits, and a web-based dashboard, provisioned with API Key(s) for programmatic access. Enterprise Fees are non-refundable except as expressly set out in the Agreement or an Order Form, and are not eligible for set off. Zep also lists other losses and interruptions it does not cover.

What Zep offers

  • Operates a cloud-based long-term memory and context-engineering software platform for AI agents and assistants (the "Platform")
  • Zep Service includes APIs, software development kits, and a web-based dashboard, provisioned with API Key(s) for programmatic access
  • Documentation (specifications, user manuals, technical materials) provided or made available, and may be modified from time to time
  • Free Plan available on a no-charge basis, plus Paid Plans (Flex, Flex Plus, Enterprise) as described on the Platform
  • For Paid Plans, Zep uses commercially reasonable efforts to provide support and to keep the Zep Service operational and available in line with industry standards
  • Self-serve subscriptions prepaid by credit card or other accepted payment method, charged at the start of each billing period; usage-based overages billed at published/Order Form rates
  • At least 30 days' prior notice before revised self-serve Fees take effect; Enterprise renewal Fee increases require 60 days' written notice
  • Enterprise invoices payable within 30 days of receipt

What they are not liable for

  • Enterprise Fees are non-refundable except as expressly set out in the Agreement or an Order Form, and are not eligible for set off
  • Fees are exclusive of taxes; Customer pays all sales, use, excise and other taxes, export/import fees, customs duties and similar charges, except taxes based on Zep's income
  • Zep reserves the right to suspend access to the Zep Service if Fees are more than fifteen (15) days overdue, in addition to other rights or remedies
  • Prohibited Data is excluded from the service, including PCI-DSS payment card data and HIPAA Protected Health Information (absent an executed Business Associate Agreement)
  • Personal Data excludes information about Customer business-contact personnel and information Zep receives independently of its relationship with Customer
  • Late payments accrue interest at 1.5% per month (or the maximum allowed by law)

Zep says it collects account and identity, contact details, product usage and analytics, device and ip data, and cookies and tracking. That data is used for registering with the Service by entering email address, user name and password.

  • Account and identity

    Collecting

    registering with the Service by entering email address, user name and password

  • Contact details

    Collecting

    help us communicate with you, respond to questions, technical support, promotional offers

  • Payment information

    Unclear

    not addressed in the policy

  • Product usage and analytics

    Collecting

    improve the Service, customize the user experience and analyze use patterns

  • Device and IP data

    Collecting

    improve the quality of the Service (browser type, device, time and date of access)

  • User content and uploads

    Unclear

    preferences you submit are collected as Non-Personal Information

  • Cookies and tracking

    Collecting

    collect Non-Personal Information and keep record of the user's preferences

  • Location

    Unclear

    general demographic information may be collected as Non-Personal Information

  • Shared with third parties

    Collecting

    shared with vendors performing services for the Company and to meet legal process or protect rights

  • Advertising and remarketing

    Unclear

    Non-Personal Information may be disclosed to partners, advertisers and other third parties

  • Used to train AI models

    Unclear

    not addressed in the policy

  • How long data is kept

    Unclear

    persistent cookies remain until deleted; data from children under 13 deleted as soon as possible

Returns and refunds

Terms

Offers a refund: No

No dedicated refund page was found. The Terms of Service state that for Enterprise tier subscriptions all Fees are non-refundable, except as expressly set out in the Agreement or an applicable Order Form, and are not eligible for set off. Self-serve subscription Fees are prepaid at the start of each billing period; no refund entitlement is stated.

  • Enterprise tier: Fees are non-refundable and not eligible for set off
  • Refunds only where expressly set out in the Agreement or an Order Form
  • Self-serve plans: Fees prepaid each billing period by credit card; revised Fees apply only to future billing periods (no refund of current-period Fees stated)

Evidence: Terms (Primary) · Privacy (Primary) · Verified Aug 22, 2026.

AI summary

Citeware analysis

Named models, written separately. These are labeled Citeware analysis, not vendor claims.

Claude Analyst

anthropic · claude-opus-5 · Aug 22, 2026

Zep is the most serious attempt yet at agent memory as infrastructure — temporal facts with validity windows, provenance back to the source episode, and published latency numbers instead of vague speed claims. The catch is that nearly every governance feature it markets is gated behind Enterprise.

As of August 2026, Zep's temporal fact invalidation is the differentiator worth paying for. The compliance story it leads with, though, only arrives on Enterprise.

The core idea is narrower and better than the marketing suggests. Most memory layers append and hope retrieval sorts it out.

Zep expires facts when new evidence contradicts them, keeps the old ones as history, and traces each one back to the episode that produced it. That lets an agent answer what a customer's plan was last March, and lets you audit why it said something.

Rebuilding that on top of a vector store is a multi-quarter project.

Strengths

  • Bitemporal model: facts carry validity windows, contradictions invalidate rather than overwrite, and point-in-time queries actually work
  • Every fact traces to its source episode, which makes agent answers auditable rather than merely plausible
  • Published p95 retrieval across graph sizes (148ms at 10K, 168ms at 100M) instead of an unfalsifiable speed claim
  • Ingest-only metering means retrieval and storage don't add cost as your agent gets chattier
  • Fuses chat history with structured business JSON into one per-user graph, which is where most homegrown memory layers fall apart
  • Three deployment trust boundaries, including Zep running inside your own VPC

Watch-outs

  • SOC 2 Type II, HIPAA BAA, audit logs and DPAs are all Enterprise-tier; the self-serve plans carry none of them
  • Flex retains API logs for a single day, which is thin for debugging a production agent
  • Observations, webhooks, analytics and custom extraction instructions all require Flex Plus or above
  • Credit burn scales with payload bytes, so high-volume event ingestion gets expensive fast
  • Accuracy and token-efficiency benchmarks are vendor-run and framed against a specific competitor
  • The privacy policy is still version 1.0 from early 2024 while the terms were refreshed in August 2026 — a small inconsistency from a vendor selling governance

GPT Analyst

openai · gpt-5.6-terra · Aug 22, 2026

Zep is a compelling enterprise memory layer for agents whose answers depend on changing customer and business facts, not merely chat recall. Its governance and temporal model are unusually well matched to production, though the ingestion-credit model deserves careful capacity planning.

As of August 2026, Zep stands out where agent memory must be temporal, governed, and auditable rather than a thin wrapper around transcript retrieval.

Fact: Zep models facts with validity windows, preserves superseded history, and can return time-specific context with source provenance. That is materially more useful than generic “memory” for support or operations agents that must explain what they knew and when.

Its published p95 retrieval figures remain under 200ms through 100M graphs, although those are vendor-reported results.

Strengths

  • Temporal invalidation preserves history while steering agents toward current facts.
  • Point-in-time queries and per-fact provenance support defensible customer and compliance workflows.
  • Published latency figures across very large graph counts are more concrete than a generic performance claim.
  • ABAC, retention controls, legal hold, audit logs, and flexible deployment make the platform credible for governed workloads.
  • Context assembly combines chat, structured JSON, and interaction data into token-conscious agent context.

Watch-outs

  • The strongest performance and accuracy claims are vendor-published, so buyers should validate them against their own data shape and retrieval workload.
  • Episode credits scale with payload size, making high-volume transcripts and verbose event payloads harder to forecast.
  • Several production-oriented capabilities, including advanced governance and private deployment, require an enterprise sales process.
  • A graph-based memory model adds data modeling and evaluation work that simpler conversational agents may not justify.

FAQ

Verified facts

Common questions about Zep.

What is Zep?

Zep Software, Inc. describes Zep as a "Context Lake" — the data-lake pattern applied to agent context, where millions of individual context graphs are governed and served as one system. Source: https://www.getzep.com/

How much does Zep cost?

Zep is freemium. It starts at $0/month. Paid plans include Free: monthly $0/mo: yearly $0/yr ($0/mo effective); Flex: monthly $125/mo: yearly $1250/yr ($104.17/mo effective, save 17%); Flex Plus: monthly $375/mo: yearly $3750/yr ($312.5/mo effective, save 17%). Check getzep.com for current prices. Source: https://www.getzep.com/pricing

Who is Zep for?

Zep is best for teams giving a production customer-support agent persistent memory of a user's prior issues and stated preferences and fusing CRM, billing and product-event JSON into a single per-user graph an agent can query. Source: https://www.getzep.com/

Should I use Zep?

Zep is best for teams giving a production customer-support agent persistent memory of a user's prior issues and stated preferences and fusing CRM, billing and product-event JSON into a single per-user graph an agent can query. Skip it if zep plan limits come from the published pricing table.. Source: https://www.getzep.com/

What are the main Zep features?

The main Zep features are Temporal context graphs that record facts with a validity window, Fact invalidation when new information contradicts the graph, with old facts kept as history, Point-in-time queries: ask what is true now or what was true on a past date, Multi-source ingest from chat history, business data (JSON) and user interactions, and Automated context assembly returning token-efficient context blocks. Source: https://www.getzep.com/

What are the downsides of Zep?

Limitations called out on this listing: Zep plan limits come from the published pricing table.; Zep packaging is recorded from the vendor pages available at analysis time.. Source: https://www.getzep.com/

What are common Zep use cases?

Common Zep use cases include Giving a production customer-support agent persistent memory of a user's prior issues and stated preferences, Fusing CRM, billing and product-event JSON into a single per-user graph an agent can query, Answering point-in-time questions such as what a customer's plan or preference was on a specific date, Reducing prompt token spend by replacing full chat transcripts with an assembled context block, Auditing why an agent said something by tracing the underlying fact back to its source episode, Detecting behavioural patterns, such as repeat upgrade timing, and feeding them to an agent as Observations, Running agent memory inside a regulated customer's own VPC via Bring Your Own Cloud, Exposing agent memory to IDEs and assistants through the Memory MCP Server. Source: https://www.getzep.com/

What data does Zep collect?

Zep collects account and identity. registering with the Service by entering email address, user name and password. Zep collects contact details. help us communicate with you, respond to questions, technical support, promotional offers. Zep does not clearly say whether it collects payment information. not addressed in the policy. Zep collects product usage and analytics. improve the Service, customize the user experience and analyze use patterns. Zep collects device and ip data. improve the quality of the Service (browser type, device, time and date of access). Zep does not clearly say whether it collects user content and uploads. preferences you submit are collected as Non-Personal Information. Zep collects cookies and tracking. collect Non-Personal Information and keep record of the user's preferences. Zep does not clearly say whether it collects location. general demographic information may be collected as Non-Personal Information. Zep collects shared with third parties. shared with vendors performing services for the Company and to meet legal process or protect rights. Zep does not clearly say whether it collects advertising and remarketing. Non-Personal Information may be disclosed to partners, advertisers and other third parties. Zep does not clearly say whether it collects used to train ai models. not addressed in the policy. Zep does not clearly say whether it collects how long data is kept. persistent cookies remain until deleted; data from children under 13 deleted as soon as possible. Source: https://www.getzep.com/legal/privacy

What should I verify before buying Zep?

Confirm pricing, limits, and support on the official site before you buy. Source: https://www.getzep.com/

Evidence: Homepage (Primary) · Verified Aug 22, 2026.

Looking for more options? Browse all Zep alternatives.

This Citeware listing for Zep was last analyzed on Aug 22, 2026. Pricing, features, and packaging are taken from the vendor site (https://www.getzep.com/) and may change. AI takes are labeled opinions from named models, not paid placement.

Source pages: Homepage · Submitted page · Pricing · About · Terms · Privacy.

Last update: monthly price; yearly price; commercial-use policy changed. Verified Aug 22, 2026.

Recommended citation: Citeware. "Zep: Pricing, Features & Alternatives." Last verified Aug 22, 2026. https://citeware.io/products/zep

To request a correction, contact [email protected].

Machine-readable: llms.txt · data.json.