TypeSafe AI
AI lab describing itself as building "machine-native intelligence infrastructure for automation,
designed to make decisions within software." Founder Diogo Almeida. No same-as: recorded — no
Wikipedia or Wikidata entry was found to verify.
Two sources now, four days apart, and the second is not TypeSafe's:
- system-one-models-and-jev (2026-09-15) — the launch post. A typed-decision model class, a first model in early access, a price, and a self-built workflow eval claiming 67.8% aggregate accuracy.
- jev-ai-sdk-guide (2026-09-19) — Vercel, reselling Jev through AI Gateway as
typesafe-ai/jev. Early access is over; the $0.042/1M input price and free output survived the transition; the limits are now documented (64k context, 32k state, ≤255 choice options, 2–10 score levels); and aconfidencestatistic appears that the launch post never mentioned.
What a distribution partner adds. Vercel is not independent — it resells the model and repeats the vendor's price — but it is the first party other than TypeSafe to describe the product, and it states two limits the launch post did not: a schema "doesn't guarantee they're correct," and calibration is a property of many answers rather than of any one. See jev-ai-sdk-guide.
Still a thin node on the lab: no funding, headcount, location or second model. Expand when a source about the company rather than the product lands.
Related
system-one-models-and-jev · jev-ai-sdk-guide · ai-gateway · inception-labs