Repolode
Submit
plastic-labs

plastic-labs/honcho

Memory infrastructure for stateful AI agents that reasons over conversations to model users, agents and groups over time.

View on GitHubWebsite
Stars
7,760
Forks
954
Open issues
174
Last commit
yesterday

What is it?

Honcho is a memory service for AI agents. You store messages and events on sessions, and Honcho processes them in the background to extract conclusions and maintain a representation of each peer, where a peer can be a human user, an AI agent, a group, a project or an idea. Applications then query that memory through a natural-language chat endpoint, session context bundles, hybrid BM25 and vector search, or low-latency static representations, and inject the results into any LLM call or framework.

The project offers a managed service at api.honcho.dev, a local stack started with the honcho-cli tool, and a self-hostable FastAPI server that needs PostgreSQL with pgvector and an LLM provider key. Python and TypeScript SDKs are provided, along with first-party memory plugins for coding agents such as Claude Code, Cursor, OpenCode and OpenClaw, and an MCP endpoint for any MCP-compatible client.

The server is licensed under AGPL-3.0. Scopes let you limit recall to a named group of sessions, and a multi-peer model records what one peer knows about another when configured.

Key features

  • 01Stores messages and events on sessions and builds per-peer representations in the background.
  • 02Natural-language chat endpoint for asking what is known about a peer.
  • 03Hybrid BM25 and vector search at workspace, session or peer level.
  • 04Prompt-ready session context with summaries, convertible for OpenAI and Anthropic message formats.
  • 05Scopes that restrict recall to a named group of sessions.
  • 06Python and TypeScript SDKs plus an MCP endpoint and coding-agent plugins.

Use cases

  • Giving a coding agent such as Claude Code or Cursor memory that persists across sessions.
  • Adding long-term user memory to a chatbot or tutoring application.
  • Keeping long conversations going by injecting summarized context into the LLM prompt.
  • Self-hosting a memory layer for an internal multi-agent system.

In collections

Open-source AI agent frameworks & appsMCP servers

Topics

#agent-memory#ai-agents#llm#memory#vector-search#context-engineering#mcp#agentic-rag#ai#ai-memory

FAQ

What is honcho?
Memory infrastructure for stateful AI agents that reasons over conversations to model users, agents and groups over time. Honcho is a memory service for AI agents. You store messages and events on sessions, and Honcho processes them in the background to extract conclusions and maintain a representation of each peer, where a peer can be a human user, an AI agent, a group, a project or an idea. Applications then query that memory through a natural-language chat endpoint, session context bundles, hybrid BM25 and vector search, or low-latency static representations, and inject the results into any LLM call or framework.
Is honcho free to use? What license does it have?
honcho is open source under the AGPL-3.0 license. Check the license terms before using it in commercial products.
Is honcho still maintained?
Actively maintained. The last commit to honcho was on Oct 23, 2026.
What can I use honcho for?
Giving a coding agent such as Claude Code or Cursor memory that persists across sessions. Adding long-term user memory to a chatbot or tutoring application. Keeping long conversations going by injecting summarized context into the LLM prompt. Self-hosting a memory layer for an internal multi-agent system.
Where can I get honcho?
The source code and installation instructions are on GitHub: https://github.com/plastic-labs/honcho

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects