Acontext

Acontext is a memory and context platform for AI agents built around skills. As an agent runs, Acontext captures the session, distills what worked into human readable Markdown skill files, and feeds those skills back on the next run, so the agent improves from its own history without embeddings or a vector database. It works with Claude Code, OpenClaw and custom agents. It is aimed at developers building production agents who need state that persists across runs. The product has four parts. Short term memory stores messages, files and artifacts across OpenAI, Anthropic and Gemini formats with automatic conversion, S3 backed disk storage and sandboxed code execution. Mid term state covers observability: task extraction from conversations, traces, success rates and dashboards. Long term skill memory creates learning spaces that turn session outcomes into reusable skills, which can be listed, read and synced to a local folder. Context engineering tools edit what an agent is shown without altering stored messages, keep prompt caches stable and produce compact session summaries. Python and TypeScript SDKs are available, and setup needs only an API key and a package install. Acontext is open source on GitHub and is also offered as a managed cloud platform with a dashboard for projects, sessions and team collaboration. The site does not publish a pricing page, so cloud costs are not listed publicly; the dashboard is where API keys are issued. The project comes from the team behind MemoDB, and the Claude Code setup is driven by a SKILL.md file that the agent reads itself.