JevPilot

# JevPilot

https://github.com/user-attachments/assets/4baef58e-54ef-4d17-9982-353a0b6e6f45

<p align="center">
  <a href="https://jevpilot.standardagents.ai">
    <img src="docs/try-jevpilot.svg" alt="Try JevPilot →" width="256" height="64" />
  </a>
</p>

A demo project showing Tesla Autopilot-like behavior using [Jev by TypeSafe AI](https://typesafe.ai/).

Sign in with Standard Agents for $0.25 of free Jev play credit. Joining the early-access list is optional.

The hosted `/api/decide` endpoint requires a valid login session. The browser sends its secure, HttpOnly session cookie; the Jev API key stays on the server.

**Interstate 08:** start in Millbrook, turn onto the signed on-ramp, merge, cruise, and exit into Cedar Town for the final stop.

## How it works

Jev receives compact tables of eligible paths, road boundaries, nearby traffic, signals, stop memory, and destination guidance. Shared table values are sent once, and instructions include only relevant situations. The road graph is sent only when choosing an alternative route after staying more than 30 meters off course for six seconds. Detailed geometry and control calculations stay local.

The simulator samples fresh steering-and-speed combinations for each decision. On the road, it favors paths that keep the whole car on asphalt. Off road, it explores a wider field of forward and reverse paths and supplies a recovery target, road boundaries, and collision predictions.

An explicit `driving_style` describes an aggressive driver: keep progressing, stop at the actual line, and close gaps before stopping behind an obstacle. Jev can choose an approach path that progressively slows to a stop 0.5 m before the line. An immediate **stop** is offered only within 2.5 m of a blocker or required stop line, at the desti