Raindrop Workshop - Open source, free, local debugger for AI agents
by•
Raindrop Workshop is the first local debugger for agents.
It's free, local, and open source.
Your local agent traces stream, token-by-token, instantly. Another Agent like Claude Code can read them over MCP. Then Claude can write evals, replay traces, fix bugs... and do it all over again.
This is the Self-Healing Agent loop. And it’s only possible on Raindrop.
Check it out and star on Github here: https://github.com/raindrop-ai/w...
Replies
hi alexis! the unsolved question with agent debuggers is the right unit of inspection. tokens are too noisy, plans are too abstract, the message/tool-call layer is where bugs actually live but the trace gets huge fast. did raindrop pick one layer as default and let you drill, or stitch all three into one view? anyway, congrats and good luck :)
Does Raindrop support custom visualizers for specific tool-call outputs?
Congrats on launch @alexisgauba ! How your solution is different from running /review, /ultrareview, /debug inside Claude Code/Codex? Am I missing something?
Finding exactly where an agent broke in a giant log file is a huge pain. Streaming traces locally is way faster than waiting on slow cloud dashboards. But the best part is definitely the MCP setup letting Claude Code read its own error logs to fix its own bugs is pretty wild.