User repository on GitHub
A show-and-tell capability study for Jev, TypeSafe's System One decision model.
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Practices & Patterns
Little Airways
A toy archipelago air-traffic demo where every in-flight judgment - divert or continue, declare an emergency, give way, and who lands first - is made live by Jev from each plane's local state in about 150 ms.
- Stars
- 6
- Language
- HTML
- License
- MIT
- Reading time
- 3 min
- Published by
- Community
- Author
- lbotinelly
- Last verified
- Sep 2026
About this project
A toy archipelago where the planes actually think. A show-and-tell capability study for Jev, TypeSafe's System One decision model: visuals by GPT-6 Astra through the Dream Loop skill, simulation and ATC logic by GLM 5.3, and every in-flight judgment made live by Jev: divert or continue, declare an emergency or keep it routine, give way, hold, even who lands first when two aircraft want the same runway.
Quoted from the project's README. Read the full README.
Works forJev for Routing
Used inJev for Gaming
Highlights
- Every ~1.6 seconds each aircraft asks four questions in one batched request: route, broadcast, response, and clearance.
- With terse option labels Jev split 0.46 give-way versus 0.41 maintain; operationally-worded criteria for the same state moved it to 0.70 versus 0.05.
- Sending comparable time estimates - fuel, fuelMinutesRemaining, destinationMinutesRemaining, and minutesToReach - made the fuel-leak drama findable.
- Safety invariants stay in code: runway occupancy is enforced deterministically, with zero simultaneous-occupancy violations across 4x-speed stress runs.
- A Cessna's emergency question returned 52/48 at confidence 0.04 while the adjacent divert question sat at a decisive 0.99.
Quickstart
bash
cd demo
export TYPESAFE_API_KEY=YOUR_API_KEY
node server.mjs # http://localhost:8765; the server proxies POST /jev to api.typesafe.aiWatch out
MIT-licensed and an independent community project not affiliated with TypeSafe AI; needs Node for the local server, which proxies POST /jev to api.typesafe.ai with a server-side key, and the findings are small experiments rather than benchmarks.
Questions about this entry
Short answers drawn from the linked project or page.
The judgments are live and the safety invariants deliberately are not. Every ~1.6 s each aircraft asks Jev four questions in one batched request — route, broadcast, response, clearance — and the tower asks it to sequence landings. Runway occupancy is enforced deterministically by the simulation with slot requests and a short-final go-around guard; which waiting aircraft lands first is the model's call.
It boots in mock mode with a hand-tuned stand-in provider and no key needed. Switching to MOCK / API pastes a key that lives in browser localStorage and travels only to your local proxy, which forwards to the TypeSafe API because that API allows no localhost CORS origins. Input is $0.042 per MTok with output free, so an evening of nine planes is single-digit cents.
The repo presents itself as a capability study and says as much about its construction: GPT-6 Astra drove the Dream Loop skill against a locked target render to produce the visuals, while GLM 5.3 wrote the simulation, ATC logic, and judgment wiring. The transferable finding is about wording, not graphics: terse option labels gave 0.46 give-way versus 0.41 maintain, operational criteria on the same state gave 0.70 versus 0.05.
That low confidence is legible and worth surfacing rather than hiding. A Cessna with a rough engine and comfortable fuel asked whether to declare an emergency came back 52/48 at confidence 0.04, while the divert question beside it sat at a decisive 0.99. The demo auto-switches to richer option wording whenever a mayday is nearby, because wording was what moved the judgment.
`cd demo` then `node server.mjs` on any recent Node with no dependencies, and open http://localhost:8765. The server proxies POST /jev to api.typesafe.ai because the API allows no localhost CORS origins. On Windows, start.bat and stop.bat do the same.
Press J to open it: every plane's current decision and every request and response rendered as raw Jev API JSON ({state, model, questions} to {answers, usage}) with latency, token counts and a running cost estimate. You can pause the feed and read it at your leisure.
MIT for the demo code. Visuals come from GPT-6 Astra through the Dream Loop skill, simulation and ATC logic from GLM 5.3, and every in-flight judgment from Jev. It is an independent community project, not affiliated with or endorsed by TypeSafe AI.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Software Bombe and live Enigma demo where Jev ranks cribs and judges plaintext.
ag_212 pointsShow HN: Jev as the Bletchley Park Analyst: A Software Bombe on EnigmaRead the thread on Hacker NewsHacker News: Show HN for public Minecraft PvP where a diffusion LM picks moves from structured state.
2 pointsShow HN: Minecraft PvP against DiffusionGemma-as-Jev, 24 ms per decisionRead the thread on Hacker NewsHacker News: Show HN: Jev finished a Pokémon Red run, with the Hall of Fame stats on the landing page.
pancomplex272 pointsShow HN: Jev Plays Pokémon RedRead the thread on Hacker NewsHacker News: Show HN: a mouse colony where each next move is Jev or a fixed rule.
1 pointShow HN: Jev-mice – mouse colony simulation using Jev and deterministic engineRead the thread on Hacker NewsMore Jev resources like this
One-pass option scoring with a local Gemma 3 4B on Apple silicon via MLX: it prefills the context once, expands that KV cache across the option batch, and scores every option in a single padded forward pass with no decoding.
Jev Practices & Patterns#community#python#open-models
Community
daseinlabs/open-jevPython
daseinlabs/open-jevPythonAn independent CUDA and vLLM take on parallel typed decisions: it scores verified single-token labels with the model's own head and assembles JSON in code.
Jev Practices & Patterns#community#python#javascript
Community
kikoncuo/jevfireJavaScript
kikoncuo/jevfireJavaScriptInteractive browser experiments that expose Jev's judgments: a support-message Decision Lab and a driving simulation where structured sensor state becomes typed steer, brake, and overtake decisions.
Jev Cookbooks#community#javascript#demo
A video comparison of local Laya against cloud Jev on a Tetris agent, where the open-weights model made decisions 11 times faster on a 16 GB MacBook Air.
Jev Practices & PatternsVideo#community#x#video
Community
Scott Chacon runs hosted Jev against two local decision models, laya and kev, in a Tetris match and a GitHub Settings filter, with a public test repo.
Jev Practices & PatternsVideo#community#video#open-models
Community
A Japanese experiment that builds a rules-based gomoku harness and replays a full Jev-versus-Jev game log to measure real per-move latency.
Jev Practices & PatternsArticle#community#article#games
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.