TypeSafe AI

TypeSafe AI is a San Francisco-based AI lab, founded in 2026 by Diogo Almeida (formerly OpenAI). The company builds "machine-native" intelligence infrastructure for automation — models that make decisions inside software rather than generating text. Its "System One" class of models is optimized for calibrated, probabilistic outputs instead of language generation. Its first model, Jev, available in early access, produces type-safe outputs with a guaranteed zero type-error rate and claims dramatically lower cost and latency than conventional LLMs on System One tasks. Its proprietary training method is RLCD (Reinforcement Learning for Calibrated Decisions).