All Posts

Projects

miStudio: Mechanistic Interpretability Workbench

An end-to-end workbench for training, understanding, and using mechanistic-interpretability sparse autoencoders (SAEs) — now with automated feature clustering, cross-layer circuit discovery, an agent-controllable MCP server, plus HuggingFace and Neuronpedia integration and …

miLLM: Mechanistic Interpretability LLM Server

An OpenAI-compatible local inference server that attaches pretrained sparse autoencoders to a running model and steers its behavior in real time — now serving whole clusters and discovered circuits, and controllable by AI agents over MCP.