TypeSafe AI: The Company Behind Jev
Last checked · Independent guide, not affiliated with TypeSafe AI
TypeSafe AI is a San Francisco startup that builds System One models, starting with Jev. It is led by CEO Diogo Almeida, a former OpenAI researcher who co-invented RLHF and InstructGPT, with COO Sasha Sheng and CTO Erik Gafni. It came out of two years in stealth on September 15, 2026 and has not disclosed its funding.
At a glance
Section titled “At a glance”| Item | Detail |
|---|---|
| Company | TypeSafe AI |
| Based in | San Francisco (office near the Embarcadero), in person five days a week |
| Product | System One models; the first is Jev |
| Public launch | September 15, 2026, after two years in stealth |
| Founders | Diogo Almeida (CEO), Sasha Sheng (COO), Erik Gafni (CTO) |
| Funding | Undisclosed. The company says it is “backed by top-tier investors” but has not named them or given amounts |
| Website and docs | typesafe.ai, docs.typesafe.ai, console.typesafe.ai, status.typesafe.ai |
The founders
Section titled “The founders”Diogo Almeida, CEO. TypeSafe’s team page credits him with co-inventing RLHF and InstructGPT, the methods behind ChatGPT and GPT-4, and lists earlier work at Google Brain. In the launch post he writes that the work at OpenAI left him asking why chat models had not led to more automation. TechCrunch reports that he left OpenAI about two years ago to start TypeSafe.
Sasha Sheng, COO. A former research engineer at Meta’s FAIR lab who worked on News Feed and AI research, according to the team page.
Erik Gafni, CTO. A repeat founder whose earlier company, Ravel, worked on multi-modal AI for DNA sequencing, and an early employee at two unicorns, Invitae and Freenome, according to the team page.
The company says its wider team comes from OpenAI, Google Brain, Meta FAIR, Stripe, Airbnb, Plaid and Docker.
What TypeSafe is trying to build
Section titled “What TypeSafe is trying to build”TypeSafe’s manifesto and documentation make one argument repeatedly: most future AI traffic will be software talking to software, not people chatting with bots, so models should behave like dependable components. The company’s phrase for it is “Build prod, not God”: narrow, testable decisions instead of one model that tries to do everything.
That shapes the product. Jev returns typed answers and calibrated probabilities instead of text, is trained with the company’s own method, RLCD, and is priced per billion input tokens with free output.
Timeline
Section titled “Timeline”| Date | Event |
|---|---|
| Sep 15, 2026 | Launch post “Introducing System One Models & Jev”; early access opens with a waitlist; Vercel AI Gateway lists Jev the same day |
| Sep 15, 2026 | The launch post draws about 1,900 points and 500 comments on Hacker News |
| Sep 17, 2026 | Last review date of TypeSafe’s “Jev 1.13 jaggedness” page, which lists known weaknesses |
| Sep 18, 2026 | OpenRouter adds typesafe/jev-1.13; TechCrunch profiles the company and reports that demand briefly overwhelmed the API |
| Sep 19, 2026 | Cloudflare adds typesafe/jev to Workers AI |
For model versions, see the Jev changelog.
What TypeSafe has not disclosed
Section titled “What TypeSafe has not disclosed”- Funding and valuation. Searches for TypeSafe AI’s funding are common, but as of September 19, 2026 the company has published no round size, investors or valuation. We will not repeat unsourced figures.
- Model details. Size, base model and training data are unpublished. See Is Jev an LLM?
- Pricing after early access. The launch post says TypeSafe expects prices to fall rather than rise, but makes no commitment.
Data and legal commitments
Section titled “Data and legal commitments”TypeSafe’s legal page links a Data Processing Agreement, a Master Customer Agreement and a Privacy Policy. The company states that it does not train its models on customer requests or responses, and offers zero data retention to enterprise customers through [email protected]. See Does Jev train on your data?
Not to be confused with Typesafe (Lightbend)
Section titled “Not to be confused with Typesafe (Lightbend)”Before 2016, the company behind the Scala programming language and the Akka toolkit was also called Typesafe; it renamed itself Lightbend. Searches for “typesafe config” or “typesafe scala” are about that older company and its libraries, not about TypeSafe AI or Jev. Programmers also use “type-safe” as an everyday term for code that the compiler checks for type errors. TypeSafe AI’s name plays on that meaning: its models’ answers always match the types you declare.
Contacting TypeSafe
Section titled “Contacting TypeSafe”This site is independent and cannot help with TypeSafe accounts. For access, billing or enterprise plans, use the console at console.typesafe.ai or the addresses in TypeSafe’s documentation ([email protected] for enterprise, [email protected] for data requests).