Circuit Tracer
Anthropic's open tools to see how AI thinks
Visit website →About
Circuit Tracer is an open-source research tool from Anthropic for studying how a language model produces a given output. It traces the model's internal computation and presents it as an attribution graph, showing which features and pathways contributed to a specific response.
Researchers can explore these graphs interactively on Neuronpedia or run the library directly on supported open models to investigate reasoning, refusals, and errors at the mechanism level.
It is aimed at interpretability researchers and technical teams working on AI transparency and safety who need to look inside a model rather than only observe its behaviour from the outside. The code and tooling are freely available.
AI-assisted draft, human-reviewed before publishing — see how we choose & review tools.
Why we picked it
Circuit Tracer stood out as a research pick, not a product one — an open-source tool that actually traces a model's internal computation into an attribution graph is worth naming for anyone trying to understand why a model produced a specific answer, not just what it produced.
Related tools
- 1 Tiden Verifies coding agents built what you asked for — specs as code Trial
- 2 GitGlow Local Git workbench with smart review for your changes and agent edits Freemium
- 3 GitHub Copilot AI coding assistant inside your editor, with a free tier Freemium
- 4 Together Link Run frontier open models in the coding harness you already use Paid
- 5 Octri Turns OpenAPI specs into docs, SDKs in 10 languages, and MCP server Freemium
- 6 FastRouter.ai One OpenAI-compatible API to route across 200+ LLMs with smart failover Freemium