Repolode
投稿する
memodb-io

memodb-io/Acontext

Open-source memory layer that turns agent task outcomes into editable Markdown skill files shared across agents and LLMs.

GitHub で見る公式サイト
スター
3,695
フォーク
332
未解決の Issue
37
最終コミット
3 か月前

これは何？

Acontext is a skill memory layer for AI agents. It captures learnings from agent sessions, such as what worked, what failed, and user preferences, and stores them as agent skill files instead of opaque embeddings. Developers can read, edit, version, and export these files as ZIP archives for use with other agents, LLMs, or frameworks.

Learning is triggered when a task is marked complete or failed. An LLM pass distills the conversation and execution trace, and a Skill Agent decides whether to update an existing skill or create a new one according to a SKILL.md schema the developer defines. Agents retrieve memory through tools like get_skill and get_skill_file, relying on tool use rather than semantic top-k search.

The project provides Python and TypeScript SDKs, a CLI for starting a self-hosted backend with Docker, and a web dashboard. It also includes a virtual persistent disk, sandboxed code execution with mountable skills, and context engineering features such as summaries and editing strategies.

主な機能

  • 01Stores learned knowledge as Markdown skill files that can be read and edited directly.
  • 02Lets developers define memory structure through a SKILL.md schema.
  • 03Retrieves memory via get_skill and get_skill_file tools instead of embedding search.
  • 04Exports skills as ZIP for use with other agents or LLMs.
  • 05Provides Python and TypeScript SDKs and a CLI for self-hosting with Docker.

ユースケース

  • Giving a coding or support agent memory of user preferences that users can inspect and correct.
  • Letting an agent learn from failed tasks and reuse successful approaches in later runs.
  • Sharing one set of skill files across agents built on different frameworks.
  • Running a self-hosted memory backend for agent development and testing.

トピック

#agent-memory#llm#ai-agent#context-engineering#agent-skills#llmops#self-hosted#agent#agent-development-kit#agent-observability

よくある質問

Acontext とは？
Open-source memory layer that turns agent task outcomes into editable Markdown skill files shared across agents and LLMs. Acontext is a skill memory layer for AI agents. It captures learnings from agent sessions, such as what worked, what failed, and user preferences, and stores them as agent skill files instead of opaque embeddings. Developers can read, edit, version, and export these files as ZIP archives for use with other agents, LLMs, or frameworks.
Acontext は無料で使えますか？ライセンスは？
Acontext は Apache-2.0 ライセンスのオープンソースです。商用製品で使う前にライセンス条件を確認してください。
Acontext は今もメンテナンスされていますか？
半年以内に更新あり。Acontext の最終コミットは 2026年7月14日 です。
Acontext は何に使えますか？
Giving a coding or support agent memory of user preferences that users can inspect and correct. Letting an agent learn from failed tasks and reuse successful approaches in later runs. Sharing one set of skill files across agents built on different frameworks. Running a self-hosted memory backend for agent development and testing.
Acontext はどこで入手できますか？
ソースコードとインストール手順は GitHub にあります：https://github.com/memodb-io/Acontext

概要はプロジェクトの README をもとに AI が生成しています。GitHub の数値は毎日更新されます。

関連プロジェクト