Harness Hub
All startups

Zep

Long-term memory for agents, built on temporal knowledge graphs.

Zep maintains durable, time-aware agent memory via its open-source Graphiti engine — facts, relationships and their history — so conversations and tasks can span sessions.

Pros & cons

Each point cites its sources and says what kind of evidence backs it. A second reviewer tried to refute every claim before it was kept.

Pros

  • MetadataBuilt on Graphiti, its Apache-2.0 temporal knowledge-graph engine with about 32k stars, which supports Neo4j, FalkorDB and Amazon Neptune.[1]
  • DocsCredits are charged only for ingested Episodes; retrieval, storage, users and threads cost 0 credits.[1]
  • DocsEnterprise offers SOC 2 Type II, a HIPAA BAA, bring-your-own-key encryption and bring-your-own-cloud deployment in your VPC.[1]
  • DocsThe free trial tier includes 10,000 credits per month for prototyping.[1]

Cons

  • DocsZep Community Edition was deprecated and moved to a legacy folder with no updates, so the full Zep service is now available only as cloud or BYOC.[1][2]
  • DocsPaid self-serve starts at $125/month (Flex), while SOC 2 Type II, the HIPAA BAA and audit logs are listed as Enterprise-only.[1]
  • DocsGraphiti needs an LLM with reliable structured output, and its README warns that other services may cause ingestion failures, especially with smaller models.[1]
  • User reportsThe Mem0/Zep LoCoMo dispute leaves neither vendor's headline benchmark score directly comparable.[1]
Best for
Agents that need temporally aware, graph-structured memory with enterprise compliance.
Watch out for
If you need fully self-hosted memory, plan to build on Graphiti yourself.

Assessed Oct 29, 2026 · adversarially reviewedSuggest a correction

People

Their open-source work

getzep/graphiti

Temporal knowledge graphs for agent memory.

Maintained by Zep

3 pros · 4 cons

Activity

Last push
· yesterday
Latest release
v0.31.2
· 4w ago
Last 90 days
6 releases
  1. v0.31.2v0.31.2 - update certifi to mitigate eMudhra compromise

    · 4w ago

    What's Changed chore(mcp): require graphiti-core 0.31.1 and set MCP version to 1.1.1 by @devin-ai-integration[bot] in https://github.com/getzep/graphiti/pull/2013 chore(ci): add the Devin bot to the CLA allowlist by @danielchalef in https://github.com/getzep/graphiti/pull/2028…

  2. v0.31.1

    · 5w ago

    What's Changed chore: set graphiti-core version to 0.31.0 for the v0.31.0 release by @devin-ai-integration[bot] in https://github.com/getzep/graphiti/pull/2010 chore: set graphiti-core version to 0.31.1 by @devin-ai-integration[bot] in…

  3. v0.31.0v0.31.0 - gpt-6-luna as the default OpenAI model

    · 5w ago

    What's Changed Publish zepai/graphiti on v.. tags like MCP by @mehulp93 in https://github.com/getzep/graphiti/pull/1862 Add pevans to the CLA allowlist [ZEPAI-3570] by @pevans in https://github.com/getzep/graphiti/pull/1883 Update root uv.lock to fix Dependabot alerts…

  4. v0.30.2v0.30.2 - FalkorDB updates

    · 2mo ago

    What's Changed fix(ci): bump gh-action-pypi-publish for metadata 2.5 by @prasmussen15 in https://github.com/getzep/graphiti/pull/1821 Bump graphiti-core to 0.30.1 by @prasmussen15 in https://github.com/getzep/graphiti/pull/1822 Fix Neo4j database routing and honor…

  5. mcp-v1.1.0mcp-v1.1.0 - Neo4j Custom Database Fix

    · 2mo ago

    ⚠️ Behavior change for existing Neo4j deployments: WHO IS AFFECTED: Self-hosted Neo4j Enterprise deployments where NEO4J_DATABASE (or database: in config.yaml) was set to something other than neo4j. Standard setups using the default neo4j database need no action. Before: The…

Checked against GitHub on Nov 21, 2026. Release notes are shortened to plain text.

On X

No posts from them in our X feed yet.

Follow @zep_ai on X · The X feed

In research (7)

Showing the 6 newest of 7 papers.

Stateful agents with memory, from the team behind MemGPT.

Letta (formerly MemGPT, out of UC Berkeley) builds an agent server with built-in memory, context management and state — letting agents persist, learn and be edited like files rather than living in a chat window.

$10M seed led by Felicis (2024)

4 pros · 3 cons

The memory layer for AI apps and agents.

Mem0 (from the EmbedChain team) provides a managed + open-source memory layer that extracts, stores and retrieves user and entity memories across sessions — a standard building block for personalization.

3 pros · 4 cons

User modeling and psychology for agent memory.

Honcho is a user-context and memory server that models what each user knows and wants, so agents can personalize without teams hand-rolling their own user-modeling stack.

4 pros · 4 cons

Connections

Relations from the Harness Hub knowledge graph, as charted in the State of Harness reports.

Maintains · 1

Similar entries · 1

Mentioned in · 1

The Harness Hub newsletter

New harness companies, projects, releases and the monthly State of Harness report, in your inbox.