Goodfire

Goodfire is an AI interpretability research company and public benefit corporation building tools to understand, inspect, and intentionally shape the internal mechanisms of neural networks. The company's core technology uses mechanistic interpretability — reverse-engineering how models represent and process information — to enable developers to debug model behavior, reduce hallucinations, and steer model outputs at a fundamental level. Goodfire offers an API and SDK that expose model 'features' (internal computational units), allowing practitioners to directly modify model behavior without retraining. Founded in 2024 by a team of researchers from Google DeepMind, OpenAI, Meta, and leading academic institutions, the company has raised over $200 million from Menlo Ventures, B Capital, and Lightspeed and is headquartered in San Francisco.