Zep

Zep is a context engineering platform that gives AI agents long-term memory by turning conversations, documents and business data into temporal knowledge graphs. It is for engineering teams building production agents that need relevant, policy-filtered facts in one low-latency call instead of stuffing whole records into the prompt.

Zep is a managed context layer for AI agents, founded in 2023 by Daniel Chalef and backed by Y Combinator (Winter 2024 batch). It started as an open-source memory store for LLM chat apps and has since repositioned, with its v3 release, as a context engineering platform. Zep ingests conversations, documents and business data, keeps a link from every fact to its source, and builds a Context Graph per user, account or domain. Facts are bi-temporal: Zep records when a fact was true and when it was ingested, so an agent can tell current information from superseded information. Retrieval combines semantic embeddings, BM25 keyword search and graph traversal and returns policy-filtered context in a single call; Zep reports retrieval under 200 ms and says returning only task-relevant facts cuts token use by about 90%. Its open-source engine, Graphiti, is Apache 2.0 licensed, has about 32,000 GitHub stars, runs on Neo4j, FalkorDB or Amazon Neptune, and ships an MCP server, though production Zep runs a proprietary graph engine called Konig. The managed service offers SOC 2 Type II, a HIPAA BAA on Enterprise, customer-managed keys and a bring-your-own-cloud option inside the customer's VPC. Named customers include Twin Health, Praktika.ai, Samsung, Quorum and HoneyBook. S&P Global Market Intelligence (451 Research) initiated coverage in April 2026, citing its temporal graph as the differentiator and cost at enterprise scale as the weakness. Its closest rival is Mem0, with which it has an open dispute over LOCOMO benchmark scores.

Category

Agent Development

Pricing

Freemium, Usage-based, Subscription

Target Market

CTOs, Enterprise Developers, AI Platform Teams, ML Engineers

Deployment

Cloud-first, API-based, Open-source

Founded

2023

Headquarters

San Francisco, USA

Team Size

1-10

• ✓
  Temporal Context Graphs

  Builds a graph per user, account or domain where every fact is dated, so agents can distinguish current facts from ones that changed.

• ✓
  Single-call hybrid retrieval

  Combines semantic search, BM25 and graph traversal to return only task-relevant facts in one request, reported at under 200 ms.

• ✓
  Attribute-based access policies

  Controls what each agent can retrieve, and logs every request, which matters when several agents share one context store.

• ✓
  Source tracing, retention and legal hold

  Links each fact to its source, expires data on a schedule and blocks deletion under legal hold, supporting audits.

• ✓
  Graphiti open-source engine

  Apache 2.0 temporal knowledge graph framework with about 32,000 GitHub stars that runs on Neo4j, FalkorDB or Amazon Neptune.

• ✓
  MCP server and multi-language SDKs

  Offers Python, TypeScript and Go SDKs and an MCP server so agents and coding tools can read and write memory directly.

• ✓
  Enterprise deployment options

  Available as managed cloud, cloud with customer-managed encryption keys, or bring-your-own-cloud inside the customer's VPC.

• •
  Personalized support agents

  A support assistant recalls a customer's prior issues, plan and preferences across sessions without replaying full transcripts into every prompt.

• •
  Health and coaching apps

  Coaching agents track how a user's goals and measurements change over time, using dated facts to avoid acting on outdated information.

• •
  Account context for sales agents

  Agents assemble a current view of an account from CRM records, documents and conversations, returning only what the task needs.

• •
  Shared memory for multi-agent systems

  Several agents read from one governed context store with per-agent access policies and a log of every retrieval.

• •
  Token cost reduction

  Teams replace long-context prompts with compact retrieved facts to lower per-request token spend while keeping answers grounded in sources.

Prices checked on getzep.com/pricing on October 19, 2026. Zep meters ingestion, not reads: one credit per episode up to 350 bytes plus one per additional 350 bytes, while retrieval, storage, users and threads are free. Flex is $125 a month for 50,000 credits and Flex Plus $375 for 200,000, with a 17% discount for annual billing. BYOC, customer-managed keys, the HIPAA BAA and audit logs are Enterprise-only and priced by quote.

This page was written from 9 sources, 6 on domains other than getzep.com.

1. 1.getzep.com — getzep.comvendor
2. 2.getzep.com — pricingvendor
3. 3.help.getzep.com — help.getzep.com
4. 4.github.com — graphiti
5. 5.ycombinator.com — zep ai
6. 6.atlan.com — zep vs mem0
7. 7.getzep.com — sp global market intelligence zep coveragevendor
8. 8.promptowl.ai — persistent memory ai agents
9. 9.hn.algolia.com — search