Circuit Tracer

Visualize LLM internal computations via attribution graphs

Visit Circuit Tracer →

Circuit Tracer is an open-source Python library by Anthropic that generates attribution graphs to help researchers understand how large language models process and route information internally. It supports use via the Neuronpedia web interface or as a standalone Python library. It is aimed at AI interpretability researchers seeking mechanistic transparency into model behavior.

Install & run

pippip install circuit-tracer

Source & docs

At a glance

Company
Anthropic
Pricing
free
API available
Yes
Self-hostable
Yes
Launched
2025-06
Last verified
2026-08-22

Capabilities

attribution-graphsmechanistic-interpretabilityllm-internals-visualizationopen-sourcepython-library

Categories

Alternatives

For AI agents: machine-readable markdown version of this page at /tools/circuit-tracer.md, or send Accept: text/markdown.