Jev is TypeSafe AI's first System One model — a frontier decision engine optimized for fast, typed, calibrated judgments (Choice / Score / Noul) instead of free-form text generation. Typical latency 70–500 ms · ~$0.042 / M input tokens · output tokens free.
Why this exists
Within 72 hours of launch the community produced browser agents booking flights in ~7 s for under $0.004, trading bots deciding every ~300 ms, computer-use agents at $0.0002 per step, plus guardrails, code reviewers, drones and games. This index helps you find the ones worth reading.
Flagship projects
- browser-use/jev-ultrafast — Zürich → London Google Flights in 7.1 s / $0.0039
- TheoLeeCJ/openjev — open re-implementation on a frozen 4B model
- tamaratran/fast-jev-compaction — Claude Code compaction → Jev keep/delete
- jarrodwatts/jev-trader — market maker with 81 ms model latency per block
- fhshaik/typesafe-mario — Super Mario from emulator RAM
- awlevin/typesafe-computer-use — macOS computer use at $0.0002 / decision
- devagrawal09/jev-review — staged code reviewer + dashboard
- RomanSlack/jev-drone — drone control in MuJoCo
Use-case catalog
- Smart workflow decisions — support triage, routing, scoring, intelligent if-statements
- Bulk map-reduce classification — 1,018 papers → 24 topics for ~$0.08
- Real-time loops & games — Doom, Wikiracing, Snake, 2048, Tetris
- Browser / desktop / mobile agents — ultrafast browser agents, Android UI, OCR computer-use
- Harness engineering & guardrails — tool-call firewalls, compaction, prompt-injection screening
- Search, reranking & RAG — relevance scoring without embeddings
- Security, moderation & safety — injection benchmarks, moderation pipelines
- Open re-implementations & research — logit reading, constrained decoding, calibration
Source
Full, always-current index: github.com/anandi1989/awesome-jev-usecases