Awesome Infra for AI › Workflow Orchestration for AI

memodb-io/Acontext

⭐ 3698 JavaScript repository created 2025-07-16

Acontext is an open-source skill memory layer designed for AI agents. Its primary function is to automatically capture and store learnings from agent runs, converting them into "agent skill files" – which are essentially Markdown files. This approach allows developers and agents to easily read, edit, and share acquired skills across different agents, LLMs, and frameworks without opaque memory structures. The project addresses the challenge of complex agent memory by making learned knowledge transparent and inspectable, enabling agents to learn from mistakes and reuse effective behaviors. Acontext emphasizes a "plain file, any framework" philosophy, using Markdown files for portability and avoiding vendor lock-in or proprietary embedding formats. Agents can retrieve skills using specific tools (e.g., `get_skill`) based on reasoning rather than semantic search. The system distills session messages, task outcomes, and execution traces via an LLM pass to infer what worked or failed, then a 'Skill Agent' updates existing skills or creates new ones based on a user-defined schema.

https://github.com/memodb-io/Acontext

AI agentsagent memoryskill managementLLM orchestrationagent learningcontext managementobservabilityself-evolving agentshuman-readable skillsMarkdown files