History & Founder
Jev didn’t appear out of nowhere. It grew out of one persistent question, two years of stealth work, and a bet that automation needs something other than a chatbot.
Diogo Almeida
Founder, TypeSafe AI
Before TypeSafe, Almeida worked at OpenAI, where he helped build the instruction-following methods that became the research behind ChatGPT. His driving question for four years: models are superhuman at chat, so where is all the automation?
“We started TypeSafe because we believe AI needs an interface software could depend on.”
— Diogo Almeida, founder’s announcement note
The road to launch
~4 years earlier
The driving question
At OpenAI, Diogo Almeida helped build the instruction-following methods behind ChatGPT — but kept asking why superhuman chat wasn't producing real automation.
2 years in stealth
A new stack
TypeSafe AI builds a new model architecture, a parallel sampler, and a new training method: Reinforcement Learning for Calibrated Decisions (RLCD).
Sep 15, 2026
Jev launches
TypeSafe releases Jev in early access — the first public 'System One Model'.
Sep 18, 2026
Ecosystem forms
jev-1.13 goes live on OpenRouter and processes 161B+ tokens in its first days. An open-source alternative, Laya, appears the same day.
Where the names come from
“System One”
From Daniel Kahneman’s Thinking, Fast and Slow. System 1 is fast and intuitive; System 2 is slow and deliberate. Jev handles the fast, intuitive judgments; LLMs do the slow reasoning.
“Jev”
Named after economist William Stanley Jevons. The Jevons paradox: making a resource cheaper increases demand for it. TypeSafe bets each drop in the cost of intelligence unlocks far more uses of it.
Context for your audience
Jev is brand new (launched September 2026) and the category — “System One Models” — is TypeSafe’s own coinage. Independent benchmarks are still early, and an open-source alternative called Laya appeared days after launch. Treat the vendor’s headline numbers as claims to verify, not settled fact.