TypeSafe AI

TypeSafe AI builds System One Models, a class of AI models designed to be used by machines for decisions inside software rather than text for people. Its first public model, Jev, returns typed outputs with calibrated probabilities and a confidence estimate for each decision, so software can act autonomously above set thresholds and send decisions to review below them. The company describes its training method as Reinforcement Learning for Calibrated Decisions (RLCD).

Jev is offered through an API console with documentation, and is aimed at developers who build automated workflows. Pricing is listed per input token, at $42 per billion input tokens. The website includes a case study in which Jack & Jill reports cutting candidate screening costs by 88% with Jev.

The company is based in San Francisco, where its listed open roles are located, spanning engineering, product, design, go-to-market and business operations. Its founders are Diogo Almeida (CEO, co-inventor of RLHF and InstructGPT), Sasha Sheng (COO) and Erik Gafni (CTO). The team includes former engineers from OpenAI, Google Brain, Meta/FAIR, Stripe and Airbnb, and the company describes itself as backed by investors.

Product: Jev, an AI model accessed via API that returns typed decisions with probabilities and confidence estimates for software automation.