Show HN: Autodidact – Self-evolving local-first AI agent
github.compip install autodidact && autodidact init I'm curious how you're handling the self-evolving part in practice. When I built browser automation agents with Claude API last year, the hardest problem wasn't the initial task execution - it was knowing when to stop iterating vs when the agent had legitimately improved its approach. Does autodidact have any safeguards against the agent just spinning in circles rewriting its own prompts, or do you rely on the user to define success criteria upfront? Yes, this is the exact engineering challenge I have been trying to solve.
Currently, the improvements only include memories and skills update, for v1.5, this is what I am trying to solve too.
Could you try the current version and let me know what you think? What happens if the model continuously hallucinates or provides wrong answers until the user give up ? Will the AI trust those wrong answers and save it ? users have option to type /wrong, then the local model will ask cloud model for the answer and learn from it.
Could you try the current version and let me know your feedbacks?