TypeSafe AI

AI lab building decision-making models for software

TypeSafe AI is a San Francisco research lab developing System One Models, a class of AI that sits inside software and makes decisions there. These models combine the reasoning ability of large language models with the speed and dependability of conventional code, with the goal of automation that holds up in production. Work began in mid-2024 on redesigning how language models are put together, with reliability and autonomy prioritised over benchmark scores, and the lab has since released Jev, its first public model. Its researchers previously worked at OpenAI, Google Brain and Meta's FAIR group, and the whole team works together in person at an office close to Embarcadero station.