TypeSafe AI

Jev is the first public release from TypeSafe AI, a San Francisco lab building what it calls System One models, a foundation model class designed for machine-to-machine automation rather than conversation. Instead of generating text token by token, Jev returns structured decisions with calibrated probabilities, which the team argues makes it a better fit for compliance pipelines, real-time decisioning, and autonomous agents that need to run without a human reviewing every output. TypeSafe describes itself as a company building a new class of large language models for true no-human-in-the-loop automation , and this launch is the first time anyone outside the waitlist can actually call the model. The pitch matters because the founding team helped build the paradigm they are now trying to move past. TypeSafe was founded in mid-2024 by Diogo Almeida, Erik Gafni and Sasha Sheng and is based out of San Francisco , with Almeida previously helping develop reinforcement learning with human feedback at OpenAI, Gafni bringing experience building production AI systems, and Sheng having worked in research engineering at Meta/FAIR . Almeida frames RLHF as the reason today's chat models remain unreliable enough to require humans in the loop, and Jev is trained with a different objective the company calls reinforcement learning for calibrated decisions, aimed at parallel rather than sequential computation. For founders and operators, the practical read is speed and unit economics on structured tasks. TypeSafe claims Jev runs roughly one hundred times faster and cheaper than a comparable LLM on decision workloads, with input priced per billion tokens, which is aimed squarely at teams whose agent stacks currently bottleneck on latency or on the cost of calling a chat model inside a tight loop. Rather than bolting guardrails onto probabilistic models after the fact, TypeSafe is embedding reliability directly into the model itself , and Jev is the first chance for developers to test whether that approach holds up outside the lab.