Acontext
An open-source AI agent memory layer that learns from task outcomes and saves reusable Markdown skills. Self-host it or use the cloud service.
Acontext turns AI agent sessions into Markdown skill files that agents can reuse on later tasks. It's for developers who want agents to retain successful approaches, past mistakes, and user preferences in memory they can inspect and correct. The full stack can run on your own infrastructure, and a hosted cloud service is also available. It's open source under Apache 2.0.
The memory stays readable. Acontext draws lessons from conversations and execution traces when tasks succeed or fail, then creates or updates files in the SKILL.md format. You can define how those files are organized, such as keeping separate memory for each project or contact. Learning runs in the background.
Agents request the skill content they need rather than relying on embedding-based similarity search. The files remain editable and portable across agents, LLMs, and frameworks. You can export them as a ZIP or sync them locally for reuse; using the hosted service means sending session messages to that service, while self-hosting puts the Acontext stack on infrastructure you control.
Acontext works with Claude Code and OpenClaw, and provides Python and TypeScript SDKs. Its integrations include OpenAI, Anthropic, LangGraph, and Agno. Alongside skill memory, it offers context compression, a persistent virtual filesystem for agents, and isolated code execution with bash and Python. Agents can access files, sandboxes, and skills through function-calling tools.
