User repository on GitHub
Three.js drone simulator with a Python backend and live TypeSafe Jev navigation
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Demos & Experiments
TypeSafe Jev Drone Demo
Three.js drone flight simulation where a typed FastAPI backend asks jev-latest for every navigation action: a Choice question selects one of nine candidate movements, an independent Noul question estimates whether the direct route is blocked.
- Stars
- 1
- Language
- Python
- Published by
- Community
- Author
- kxzk
- Last verified
- Oct 2026
About this project
A live Three.js drone simulation with a fully typed Python/FastAPI backend. TypeSafe's jev-latest chooses each navigation action; its actual probabilities, confidence, latency, model input, and output are visible beside the flight. Requires Node.js 20.19+ (or 22.12+) and uv. Set TYPESAFEAPIKEY in .env. Environment files are ignored by Git; only the empty .env.example is tracked. The key is never sent to the browser.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- Python
Highlights
- Python enumerates nine candidate movements with swept collision clearance, predicted endpoint and progress toward the checkpoint; Jev only chooses between them.
- The chosen movement is revalidated against the world, and the drone holds if it is blocked or confidence is below 0.12, never a heuristic fallback.
- A live Choice distribution, the Noul obstruction estimate and a contained Inspect dialog show exactly what Jev received and returned for each step.
- Auth, timeout and malformed-response failures pause the flight with an error, overload retries twice with backoff, and a session caps at 150 requests.
Watch out
Jev receives numeric and textual state, never images or video, and the README is explicit that this is a simulation with perfect geometric observations and simplified motion, not a real aircraft controller. The confidence shown is the API's distribution statistic, not a calibrated probability of safe flight; the Noul estimate is displayed for inspection while deterministic geometry is what actually enforces collision checks. Hovering between model calls is part of the design, not a bug. There is no offline fake-model mode, so development needs a real key, and the request budget is per browser tab: flights spend real credits while running, resetting does not clear the 150-request counter, and the check_live scripts fly complete missions against the real API. The server binds to loopback and allows local browser origins, so remote deployment and real drone integration are explicitly out of scope. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
A simulation, and the README says so up front. Jev gets numeric and textual state, not images or video, the geometry observations are perfect and the motion is simplified, and the server binds to loopback while remote deployment and real drone integration are named as outside the demo's scope. What transfers is the pattern: enumerate valid actions in code, let the model choose one, then re-verify the choice before acting.
Not as a safety claim. The README states plainly that confidence is the API's distribution statistic, not a calibrated probability of safe flight, and that the Noul obstruction estimate is shown for inspection only. Safety comes from deterministic geometry: Python revalidates the chosen movement against the world and holds if it is blocked or if confidence falls below 0.12, so a confidently wrong answer still cannot pass a collision check.
Authentication, timeout and malformed-response failures pause the flight and surface an error, because there is no fake-model mode to fall back on. Rate limit and overload conditions retry twice with backoff, and six consecutive holds pause a flight as stalled. Each connected session allows at most 150 navigation requests, retries can add HTTP calls, and resetting does not clear that counter, so long sessions stop rather than run unbounded.
The offline checks are free: uv run pytest -q, uv run mypy backend scripts tests, uv run ruff check backend scripts tests, npm run build and npm test. The two scripts that are not free are uv run python -m scripts.check_live and the same command with --barrier, which fly complete missions through the real API and spend credits. Frontend contract tests consume Python-serialized fixtures so the two languages cannot drift silently.
Python predicts nine short candidate movements with swept collision clearance, then a Choice question selects the next action and an independent Noul estimates whether the direct route is blocked.
Each connected flight session allows at most 150 navigation requests, retries can add HTTP calls, and resetting the course does not clear that counter.
Requires Node.js 20.19+ and uv: run uv sync and npm install, set TYPESAFE_API_KEY in .env, build, then serve with uvicorn on 127.0.0.1:8000 and select Start flight.
More Jev resources like this
46GitHub stars
Open 150M encoder that reproduces the typed noul, choice, and score interface in one forward pass, trainable on a free GPU and runnable locally for nothing.
Jev Demos & Experiments#community#jevbest#catalog-import
941GitHub stars
Library that adds schema-guaranteed typed outputs to existing autoregressive LLMs without changing weights, served through SGLang or a hosted API.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
TypeLLM/TypeLLMPython
TypeLLM/TypeLLMPython45GitHub stars
Drop-in service that exposes Upstage Solar models through the System One shape, accepting state plus noul, choice, or score questions.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
hunkim/solar-mini4-jevPython
hunkim/solar-mini4-jevPython38GitHub stars
Research recipe that fuses two frozen open readers into Jev-style yes/no judgments with a fit-free router, confidence gate, and candidate sets.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
FeiLiuEM/open-medical-jevPython
FeiLiuEM/open-medical-jevPython62GitHub stars
Simulation study that drives one xArm7 apple-to-plate task with Jev, a large reasoning model, and a small model through identical incremental controls.
Jev Demos & Experiments#community#jevbest#catalog-import
40GitHub stars
Open research toolkit that encodes an image once and answers multiple typed choice, noul, and score questions from a shared distribution, plus a local 0.6B text-decision model.
Jev Demos & Experiments#community#jevbest#catalog-import
Community
IamBusy/OpenJev-VisionPython
IamBusy/OpenJev-VisionPythonFrom the Jev guides
Original write-ups on what this entry is for.
A game tick is a state plus a legal action list. How Clash Royale, Minecraft, and the earlier Mario, Pokémon, StarCraft, and drone loops ask Jev one step at a time.
Jev Demos & Experiments#gaming#agents#control
Read article
From the Jev community
Posts from builders shipping with Jev right now.