# Raindrop Workshop

> A free, local, open-source debugger for AI agents that streams every token, tool call, and decision in real time and integrates with coding agents to write and run evals.

Raindrop Workshop is a local debugging and evaluation tool for AI agents, built by Raindrop AI and released under the MIT license. It installs via a single curl command and runs entirely on your machine at `localhost:5899`, streaming agent traces in real time without polling or cloud dependencies. The project launched in May 2026 and quickly attracted attention from agent developers, accumulating 686 GitHub stars within days of launch according to the repository.

## What It Is

Workshop is a local observability and eval-automation tool designed specifically for developers building AI agents. It captures every token, tool call, and span from a running agent and surfaces them in a browser-based UI. Beyond passive observation, it integrates with coding agents — primarily Claude Code — to close a self-healing loop: Claude reads the traces, writes evaluation assertions against your codebase, runs the agent, sees failures, patches the code, and re-runs until all assertions pass.

## How the Self-Healing Eval Loop Works

The core workflow has three stages:

- **Instrument:** Run `/instrument-agent` inside your coding agent to inject Raindrop tracing into your project and open Workshop in the browser.
- **Observe:** Traces stream live into the span tree view, showing token counts, tool call durations, costs, and system/user/assistant message content.
- **Fix:** Claude Code reads the trace, generates a typed eval file (e.g., `test_symptom_followup.ts`), runs the agent against it, identifies failing assertions, patches the source, and re-runs — repeating until every assertion passes.

The `/setup-agent-replay` command scaffolds an HTTP endpoint that replays a production trace against your local agent code, enabling regression testing against real-world inputs.

## Compatibility and Integrations

Workshop is designed to work across the major agent frameworks and language ecosystems without lock-in:

- **Languages:** TypeScript, Python, Go, Rust
- **Agent SDKs:** Vercel AI SDK, OpenAI Agents SDK, Anthropic SDK, LangChain, LangGraph, CrewAI, Mastra, Pydantic AI, DSPy, Google ADK, Strands, Agno, Deep Agents
- **Cloud providers:** AWS Bedrock, Azure OpenAI, Vertex AI
- **Coding agents:** Claude Code, Codex, Devin, Cursor, OpenCode

## Deployment Model

Workshop runs as a local daemon with a Vite-based browser UI. All trace data is stored in a local SQLite database (`~/.raindrop/raindrop_workshop.db` by default). No data leaves the machine unless the developer explicitly configures otherwise. The CLI exposes commands for starting, resetting, and checking the health of the daemon. Configuration is handled via environment variables for port, database path, and SDK-side trace mirroring.

## Update: v0.1.8

The latest release is v0.1.8, published on May 19, 2026. The repository was created on May 1, 2026, making this an early-stage but actively iterated project. The GitHub repository shows 686 stars and 23 forks as of the last update on May 20, 2026, with zero open issues. The project's rapid iteration cadence and community response on X — with builders describing it as immediately useful for spans, token counts, cost visibility, and timing — signal active development momentum.

## Features
- Live streamed traces — every token, tool call, and span streams in real time
- Coding-agent integration with Claude Code, Codex, Devin, Cursor, and OpenCode
- Self-healing eval loop: Claude writes evals, runs agent, fixes code, re-runs until passing
- Local replay via /setup-agent-replay scaffolding an HTTP endpoint
- Browser-based span tree UI at localhost:5899
- Local SQLite storage — no cloud dependency
- Single-line install via curl
- CLI commands: start, setup, status, reset, update
- Compatible with TypeScript, Python, Go, and Rust
- Supports AWS Bedrock, Azure OpenAI, and Vertex AI providers

## Integrations
Vercel AI SDK, OpenAI Agents SDK, Anthropic SDK, LangChain, LangGraph, CrewAI, Mastra, Pydantic AI, DSPy, Google ADK, Strands, Agno, Deep Agents, Claude Code, Codex, Devin, Cursor, OpenCode, AWS Bedrock, Azure OpenAI, Vertex AI, MCP, Temporal

## Platforms
WINDOWS, MACOS, LINUX, WEB, API, CLI

## Pricing
Open Source

## Version
v0.1.8

## Links
- Website: https://www.raindrop.ai/workshop/
- Documentation: https://www.raindrop.ai/docs/workshop/overview
- Repository: https://github.com/raindrop-ai/workshop
- EveryDev.ai: https://www.everydev.ai/tools/raindrop-workshop