It’s designed to plug into frameworks like CrewAI, AutoGen, or LangChain and help agents learn from both successful and failed interactions - so instead of each execution being isolated, the system builds up knowledge about what actually works in specific scenarios and applies that as contextual guidance next time. The aim is to move beyond static prompts and manual tweaks by letting agents improve continuously from their own runs.
Currently also working on an MCP interface to it, so people can easily try it in e.g. Cursor.
It’s designed to plug into frameworks like CrewAI, AutoGen, or LangChain and help agents learn from both successful and failed interactions - so instead of each execution being isolated, the system builds up knowledge about what actually works in specific scenarios and applies that as contextual guidance next time. The aim is to move beyond static prompts and manual tweaks by letting agents improve continuously from their own runs.
Currently also working on an MCP interface to it, so people can easily try it in e.g. Cursor.