Observatory
See the model thinking, feature by feature, token by token.
The narrative layer Neuronpedia lacks. Scrub a prompt through the residual stream, watch features ignite, click into circuits, compare reasoning across models.
Tools
Trace Theater
LiveCinematic timeline. Scrub token-by-token, features ignite in real-time, intervention slider with live counterfactuals. Exportable as video.
Circuit Canvas
LiveFigma-style attribution graphs. Nodes = features, edges = strength. Click a node, see the circuit that produced the output. AtP* + Sparse Feature Circuits pipeline.
Atlas
Q2Google Scholar for features. Cross-model semantic search: find "overconfidence" and see matches across Qwen, Gemma, Llama, Claude. The Rosetta Stone.
Compare
Q2N-way diff. Same prompt in four models side-by-side; or four prompts in one model. Heatmap of divergence. Killer for reasoning-model evaluation.
Backing SAEs
Every trace, circuit, and atlas entry is grounded in a publicly-trained sparse autoencoder. All artifacts on HuggingFace, all training recipes reproducible.