raindrop-ai/workshop

Give your coding agent the power to write and run agent evals.

What it solves

Raindrop Workshop is a local debugger for AI agents. It allows developers to see exactly what is happening inside an agent's thought process in real-time, eliminating the guesswork involved in debugging agentic workflows by providing visibility into every token, tool call, and decision.

How it works

Workshop instruments your agent by mirroring traces to a local daemon and a browser-based UI. It integrates with coding agents like Claude Code to enable a self-healing loop where the agent can read its own traces, write evaluations (evals) against the codebase, and fix bugs until assertions pass. It also provides a local replay feature to simulate production traces against local agent code.

Who it’s for

Developers building AI agents using various SDKs (such as LangChain, CrewAI, or Pydantic AI) and coding agents (such as Cursor, Devin, or Claude Code) across multiple languages including TypeScript, Python, Go, and Rust.

Highlights

  • Live streamed traces: Real-time visibility into tokens and tool calls without polling.
  • Self-healing eval loop: Integration with Claude Code to automatically write and fix code based on trace failures.
  • Local replay: Ability to scaffold HTTP endpoints to replay production traces locally.
  • Broad compatibility: Supports a wide range of AI SDKs, providers (AWS Bedrock, Azure OpenAI, Vertex AI), and coding agents.

Related

  • Project
    liaohch3/claude-tap

    A local proxy and trace viewer for AI coding agents that allows users to inspect API traffic, system prompts, and tool calls to debug agent behavior.

  • Project
    rajudandigam/agent-inspect

    A local-first observability and debugging tool for TypeScript AI agents that turns execution traces into readable trees and deterministic trajectory checks for CI.

  • Project
    simple10/agents-observe

    A real-time observability dashboard for Claude Code and Codex agents that provides visibility into subagent hierarchies, tool calls, and token usage.

  • Project
    traceroot-ai/traceroot

    An open-source observability and self-improving layer for AI agents that detects production failures, root-causes them to source code, and opens verified fix PRs.

  • Project
    vercel-labs/dev3000

    An agent-first local web debugging runtime that provides a unified timeline of server logs, browser errors, and network activity for AI coding agents.