Acontext

Why we included this project

Any team that has watched an agent repeat the same costly mistake knows why durable memory matters. Acontext's answer is to treat memory as files: it watches agent runs, pulls the lessons from both completed and failed tasks, and writes them into ordinary Markdown files you can edit by hand and commit to git like any other code. Because the memory lives in plain text, it is not a black box: you can read exactly what the agent picked up, correct it by editing the file, and reuse the same skills in Claude, LangGraph, the AI SDK, or any tooling that reads files. The design skips embedding vectors and semantic search in favor of letting the agent pull the skills it needs on demand, which keeps things simple and portable. If you would rather not depend on the hosted service, there is an API server you can self-host.