Observability startup Raindrop AI launched Workshop on Friday, an open source, MIT-licensed tool that gives developers a local debugger built specifically for AI agents. It runs as a background daemon and streams every token, tool call, and decision to a dashboard at localhost:5899 in real time. Everything is stored in a single lightweight SQL database file, keeping memory usage low and data off external servers.
The tool's headline feature is what Raindrop calls a self-healing eval loop. Coding agents like Claude Code can read the captured traces, write evaluations against the codebase, spot the logic error, and re-run the agent until all assertions pass. Raindrop co-founder and CTO Ben Hylak, formerly of Apple and SpaceX, said the tool was built to give developers a sane way to debug agents locally. In a practical example, if a veterinary assistant agent skips necessary follow-up questions, Workshop captures the full trajectory and hands it off to the coding agent to fix autonomously.
Workshop supports TypeScript, Python, Rust, and Go, and integrates with the Vercel AI SDK, OpenAI, Anthropic, LangChain, LlamaIndex, and CrewAI. It works with Claude Code, Cursor, Devin, and OpenCode. Installation takes a single shell command on macOS, Linux, or Windows. The source code is on GitHub and uses the Bun runtime. The MIT license means it stays free for everyone, including enterprise teams who want full control over their data.
