Acontext

An open-source AI agent memory layer that turns task outcomes into editable Markdown skills, with self-hosted and cloud deployment options.

Acontext gives AI agents reusable memory from past tasks. It's for developers who want agents to retain successful approaches, mistakes, and user preferences in files they can inspect and correct.

Learning from outcomes: Capture lessons from conversations and execution traces when tasks succeed or fail, then create or update Markdown skill files in the background.
Editable memory: Read and change the stored skills, and define their organization around projects, contacts, or another structure that fits your agent.
Selective recall: Let agents request the skill content they need through tools, without embedding-based similarity search.
Portable skills: Export skills as a ZIP or sync them locally for reuse across agents, LLMs, and frameworks.

Acontext works with Claude Code and OpenClaw. Python and TypeScript SDKs support integration with OpenAI, Anthropic, LangGraph, and Agno. It also provides context compression, a persistent virtual filesystem, and isolated bash and Python execution for agents.

The full stack can run on your own infrastructure, and a hosted cloud service is available. With the cloud service, session messages go to Acontext for learning; downloaded skill files can then be used locally. Acontext is open source under Apache 2.0.