# Dyno Lab

A free, open-source Mac app for AI safety and interpretability research, letting a user inspect a model's internal activations, run probes and sparse autoencoders, and test a hypothesis interactively.

Dyno Lab is a Mac app for running AI models locally and investigating how they behave, aimed at AI safety, alignment, and interpretability research. In one visual workspace, a user can send a prompt, inspect the request, capture the model's internal activations, and test a hypothesis about its behavior, saving and comparing results across experiments through a notebook that keeps a question, prompts, results, and observations together.

Larger GGUF models can run through an experimental local GPU pool that connects a Mac to a Windows NVIDIA worker over the network. Token limits and incomplete runs are shown explicitly rather than silently treated as complete answers. It bundles Python and MLX, ships a Python SDK, an HTTP API, and MCP support for automating experiments, and requires Apple Silicon and macOS 14 or later; model weights are downloaded separately after installing.

Website: https://dynolab.dev