Raindrop AI has released Workshop, an open-source MIT-licensed tool that enables local debugging and evaluation of AI agents via a single .db file and a local dashboard accessible at localhost:5899.

The tool supports multiple languages including TypeScript, Python, Rust, and Go, and integrates with popular SDKs and frameworks such as OpenAI, Anthropic, LangChain, LlamaIndex, Vercel AI SDK, Claude Code, Cursor, Devin, and OpenCode.

A standout feature is the self-healing eval loop, where agents can read traces, generate and apply evaluations to fix code, and re-run until assertions pass.

Raindrop promotes community engagement with limited-edition merchandise for users who install and run a specific command.

MIT licensing keeps the tool free and open-source, supports data sovereignty for enterprises, and encourages ongoing community contributions.

Workshop runs on macOS, Linux, and Windows, is installable with a single shell command, and its source is available on GitHub with Bun runtime support.

To address privacy, Workshop streams every token, tool call, and decision to a local UI, keeping traces entirely on the developer’s machine.