TypeSafe AI
TypeSafe AI is the company that trains and hosts Jev, the first public System One decision model. This page is company context for builders — waitlist, keys, DPA questions, “who is behind RLCD” — not a careers microsite and not TypeSafe’s about page.
Official team, culture, and open roles: typesafe.ai/team. Official product docs: docs.typesafe.ai. This guide: about jev.pro. Founders in one place: Jev founders.
What TypeSafe is (public description)
TypeSafe’s own materials describe a lab building machine-native, composable intelligence: unstructured state in, typed probabilistic decisions out. The 15 September 2026 launch post introduces System One models and Jev as the first public instance — a function-call shape, not a chatbot.
Public product surface builders actually use (all their properties):
- HTTP
POST /v1/systemoneandGET /v1/models - Official Python and JavaScript SDKs, Playground, agent skill
- Models card: Jev 1.13 pricing, aliases, context, data-handling pointers
We do not restate unpublished architecture. TechCrunch (18 September 2026) notes Almeida is tight-lipped and that outside observers suspect an open-weight base — that remains speculation, not a fact on this page.
Public company facts (dated)
From TypeSafe’s team page and the 15 September 2026 BusinessWire seed announcement (retrieved 2026-09-21):
- Founded 2024 (BusinessWire)
- Headquarters San Francisco; team page: in-person five days a week near Embarcadero
- Seed: approximately $40 million, led by DCVC (BusinessWire, 15 September 2026)
- Founders: Diogo Almeida (CEO), Erik Gafni (CTO), Sasha Sheng (COO)
- Team page also says people from OpenAI, Google Brain, Meta/FAIR, Stripe, Airbnb, Plaid, Docker, “and more,” and that they are “backed by top-tier investors”
TechCrunch: Almeida left OpenAI “two years ago” (relative to 18 September 2026) to start TypeSafe; Jev launched that week in early access.
We will not invent valuation, headcount, or a customer logo wall.
Why builders should care (not a hagiography)
Company context matters for mental models, not for celebrity:
- RLCD (Reinforcement Learning for Calibrated Decisions) is TypeSafe’s named training story. If you treat Jev like an RLHF chat model, you will ask it to write. RLCD explainer.
- System One is their product class name (Kahneman-inspired, their FAQ). Operationally: snap judgments in code. What is a System One model?
- Procurement: hosted model, no customer LoRA in public docs, ZDR/DPA on TypeSafe legal pages, higher limits via
[email protected]. Jev for enterprises. - Access: early access / waitlist is their queue. Get access.
TypeSafe vs this site
| Surface | Who |
|---|---|
| typesafe.ai, docs.typesafe.ai, console | TypeSafe AI |
| jev.pro | Independent unofficial field guide |
We do not issue keys, staff their waitlist, or speak for sales. Affiliation lives in the site footer.
FAQ
Is TypeSafe a “frontier lab”?
Almeida told TechCrunch he would like the main product to be intelligence, and contrasted TypeSafe with labs whose “main product” is fear or hype. That is his quote, not our taxonomy.
Where is the official team page?
https://typesafe.ai/team — always prefer it over this summary.
Does jev.pro partner with TypeSafe?
No.
What this page does not claim
- No invented customers, revenue, or GPU-cluster sizes.
- No endorsement of the seed announcement.
- Not a clone of a rival
/typesafe-ai/handbook page.