TypeSafe AI (company)

TL;DR TypeSafe AI, Inc. is a San Francisco AI lab building "machine-native, composable AI": models that return typed decisions instead of text. It emerged from stealth on 2026-09-15 with ~$40M in seed funding led by DCVC and opened early access to Jev. Its product surface is one HTTP endpoint (POST /v1/systemone), a web console, Python and JavaScript SDKs, an agent skill, and an open-source LLM adapter.

Facts

What the company says it is doing

The homepage frames TypeSafe as "the opposite research direction" to RLHF. Its argument, verbatim in outline: RLHF produced LLMs optimized for human preference — "superhuman at instruction following," what we now call "chat" — but with "mode dropping, overconfidence, and lack of reliability," so LLMs "require humans-in-the-loop." TypeSafe says it built "a new class of models, System One Models, to be natively used by machines… a new architecture, a new sampler, and a new training algorithm: Reinforcement Learning for Calibrated Decisions (RLCD)." Its three homepage taglines are "Decisions, not strings," "calibrated confidence," and "more like code."

The company's mission statement lives in Manifesto: Composable AI — Build Prod, Not God ("Composable AI: Build Prod, Not God"). The launch argument is in Blog: Introducing System One Models & Jev (2026-09-15).

Products

Homepage claims (attributed, not verified here)

All of the following are TypeSafe's own marketing claims from https://typesafe.ai:

• "193.6x Faster, 444.6x Cheaper." Footnoted on the homepage as "*based on workflows for System One tasks (proof)". The launch blog says this figure "comes from" the workflow evals site and that "we expect that these are on the higher end of real world gains."
• Side-by-side counter: "TypeSafe AI — Cost $0.000081, Completed in 0.114s" versus "LLMs — Cost $0.013880, Completed in 8.566s."
• "$42 Per Billion input tokens" and "238x Lower input price than Claude Fable 5.1." The per-token price is confirmed in the docs: $0.042 / MTok input, output free (Models, aliases, pricing, rate limits, context).
• "Zero Hallucinations", qualified on the same page by "Every Jev decision comes with a confidence estimate, so your software can act when confidence is high and escalate when it is not." The blog is narrower: schema conformance is guaranteed and its 0% figure "is not empirical."
• "Jev's intelligence per dollar is literally off the charts" (chart: "Workflow Intelligence vs. Cost").
• "Machine-Native Intelligence. LLMs produce words for people. Jev produces typed decisions and is more like code: reliable, fast, self-consistent, and type-safe."

Third-party commentary pushes back on several of these; see Press and third-party coverage.

FAQ questions listed on the homepage

Only the first answer is captured in the raw text; the rest are collapsed accordions.

1. What are System One Models? What is Jev? — "System One Models are a new class of AI model built for decisions inside software. Jev is TypeSafe's first public System One Model, optimized for automation. Send Jev structured questions and get typed decisions with probabilities and confidence that your software can act on."
2. Is Jev just a smaller LLM?
3. How is this different from JSON mode or structured outputs?
4. How can Jev be so fast and inexpensive?
5. Can you make Jev even faster?
6. Are these prices temporary or subsidized?
7. What is Jev good at? Where does it struggle?
8. Can Jev still get things wrong?
9. Is Jev deterministic?
10. How do I get started or ask a question?

Funding and launch

Per the Business Wire release carried by Yahoo Finance (2026-09-15, marked "This is a paid press release"): TypeSafe AI "emerged from stealth with $40 million in seed funding led by DCVC," founded by "former OpenAI researcher and co-inventor of RLHF/ChatGPT, Diogo Almeida, with Erik Gafni and Sasha Sheng." James Hardiman, General Partner at DCVC, is quoted. DCVC is the only investor named in the captured press. The team page adds only that TypeSafe is "backed by top-tier investors."

The release also claims Jev "delivers frontier-level intelligence at less than 100 milliseconds of latency and is up to 100 times faster and less expensive than other frontier models" — note this conflicts with the docs and blog figure of 70–500 ms end-to-end and with the homepage's 193.6x/444.6x multipliers.

People and culture

See Founders and team for the founders' bios. The team page says the team comes "from OpenAI, Google Brain, Meta/FAIR, Stripe, Airbnb, Plaid, Docker, and more," works in person five days a week in San Francisco, and lists five values (Positive-Sum Games with Long-Term People; Thinking in Bets; First Principles; Prioritize Learning; Passion).

Relationship to Jev

TypeSafe AI builds, hosts, and sells Jev. Jev is not fine-tuned per customer: "the same weights serve every account" and TypeSafe states it does not train on customer requests or responses (raw/docs/models.md). Commercial terms are the Master Customer Agreement, which defines the console as the "Web Interface" and the hosted API as the "API," together the "Services" — see Legal: MCA, DPA, privacy, data retention.

Related

• Jev (model) — the model itself
• Manifesto: Composable AI — Build Prod, Not God — the company's stated mission and argument
• Blog: Introducing System One Models & Jev (2026-09-15) — the launch post
• Press and third-party coverage — third-party coverage and its caveats
• Founders and team — founders and culture
• typesafe-ai GitHub organisation and repos — the open-source surface
• System One Models — what a System One model is

Sources

• raw/site/typesafe-ai-home.txt (https://typesafe.ai/)
• raw/site/team.txt (https://typesafe.ai/team)
• raw/site/blog-introducing-system-one.txt (https://typesafe.ai/blog/introducing-system-one-models-and-jev)
• raw/site/press-yahoo-funding.txt (https://finance.yahoo.com/technology/ai/articles/typesafe-ai-emerges-stealth-40m-190000776.html)
• raw/site/press-theregister.txt (https://www.theregister.com/ai-and-ml/2026/09/16/typesafe-ai-debuts-model-for-machines-that-plays-doom/5296711)
• raw/site/press-developersdigest.txt (https://www.developersdigest.tech/blog/typesafe-jev-system-one-models-release-guide-2026)
• raw/site/press-substack-maio.txt (https://anthonymaio.substack.com/p/jev-the-language-model-that-wont)
• raw/site/press-devto.txt (https://dev.to/valyuai/how-to-use-jev-a-practical-guide-to-typesafes-system-one-model-g5e)
• raw/site/typesafe-ai-legal_mca.txt (https://typesafe.ai/legal/mca)