Repository navigation
Home
Tracedown finds the step that caused your AI agent's failure. Your tracing tool shows what the agent did. Tracedown tells you which step broke it, shows the evidence, suggests a fix and proves whether it works. It is open source (Apache-2.0), and every feature is free.
Think of a car mechanic. The warning light on the dashboard is the symptom. A good mechanic doesn't replace the light; they follow the wires back to the part that failed. Tracedown does the same with an agent run: it starts where the failure showed up and follows the data back to the step that caused it.
Try the live demo: eight demo failures, read-only, nothing to install. The free host sleeps when idle, so the first visit can take a minute.
| I want to… | Go to |
|---|---|
| Use Tracedown on my agents | The documentation site: quick start, importing traces, integrations, configuration |
| Understand it in plain words | How Tracedown Works and the Glossary |
| Help build it | Your First Contribution, then the Architecture Tour |
The documentation site is the user manual. This wiki is for understanding the project and working on it: how the code fits together, how to add to it, why it was built this way, and what comes next.
Understand
- How Tracedown Works: the method, in pictures
- Glossary: run, span, step, symptom, root cause, replay…
- Accuracy and Benchmarks: how often it is right, and how we measure it
- Design Decisions: why rules first, why no percentages, why a stdlib-only SDK
Contribute
- Your First Contribution: from picking an issue to a merged pull request
- Development Setup: Codespaces or your own machine
- Architecture Tour: where everything lives in the code
- Writing a Detector: teach Tracedown a new kind of failure
- Adding a Framework Integration: support a new agent framework
- Adding a Demo Story: a small agent that fails in a realistic way
- Reporting a Wrong Diagnosis: the most useful report you can send
Run and maintain
- Troubleshooting: common errors and their fixes
- Release Process: how a version reaches PyPI and npm
- Roadmap: what is done, in progress and next
- Community and Support: where to ask, and how to say hello
| Install | pip install agent-tracedown-server && tracedown up |
| Packages |
agent-tracedown (Python SDK), agent-tracedown-server, agent-tracedown on npm
|
| Works with | LangGraph, CrewAI, OpenAI Agents, Claude Agent SDK, Google ADK, Pydantic AI, AutoGen, Semantic Kernel, LlamaIndex, MCP, Vercel AI SDK; imports from Langfuse and OpenTelemetry |
| License | Apache-2.0, no paid edition |
| Source | https://github.com/nbn07/agent-tracedown |
Tracedown is Apache-2.0, every feature free · Repository · Documentation · Something out of date? These pages live in wiki/: open a pull request or an issue.
Understand
Contribute
- Your First Contribution
- Development Setup
- Architecture Tour
- Writing a Detector
- Adding a Framework Integration
- Adding a Demo Story
- Reporting a Wrong Diagnosis
Run and maintain
