Acontext

Acontext provides a skill memory layer for AI agents, enabling them to learn from previous runs by capturing sessions and distilling successful outcomes into reusable skill files. These skill files are written in Markdown format, making them human-readable and portable for use in different agent frameworks. The tool is designed to support continuous agent improvement by feeding learned skills back into agents for future tasks, forming a loop of learning and reuse.

The platform captures agent sessions, stores messages and artifacts, and then processes these sessions to extract skills based on successful task outcomes. md schema, which agents can later recall using specific functions. Users can list, read, and download learned skills to a local directory, and export them as ZIP files for use in any framework. Acontext does not use embeddings, focusing instead on transparent, editable, and portable skill representations.

Acontext offers SDKs and integrations for frameworks such as OpenAI, Anthropic, LangGraph, and Agno, supporting both Python and TypeScript environments. The tool can be accessed via an API key and installed using pip, with commands available for creating learning spaces, attaching sessions, and managing skill files. It is suitable for developers and teams building or operating AI agents who require persistent and portable skill memory across different runs and environments.

The platform is open-source and can be self-hosted on user infrastructure with a single command. Acontext is positioned as a tool for enhancing agent autonomy by providing a structured memory layer that agents can read, edit, and move between systems.