pi-jev

# pi-jev

Speed-focused [Pi](https://pi.dev) extension combining a local coding model with hosted [TypeSafe Jev](https://docs.typesafe.ai/concepts/how-to-build-with-system-one) judgments. File retrieval reduces context sent to the generator; exact-result caching and an outage cooldown bound repeated network overhead. Qwen, DeepSeek, and other generators continue to write code and use Pi's tools.

**This does not accelerate model inference.** It targets end-to-end latency by retrieving focused excerpts and avoiding redundant hosted calls. Every uncached Jev call adds latency. Reading small files directly is often cheaper. See the retained positive and negative measurements in [VALIDATION.md](VALIDATION.md).

## Install into Pi

`pi-jev` is a Pi package that exposes one extension (`./src/index.ts`). Full instructions live in [docs/install.md](docs/install.md).

```sh
npm ci                              # install test/benchmark dependencies

# Try it for one session, nothing written to settings:
pi -e ./src/index.ts --model opencode-go/deepseek-v4.1-flash

# Register this checkout persistently (global settings):
pi install "$(pwd)"

# Or for one project only (writes .pi/settings.json):
pi install -l /absolute/path/to/pi-jev

# From git, pinned to a tag or commit:
pi install git:github.com/madeye/pi-jev@v0.1.0
```

Verify and manage the install with `pi list`, `pi config` (enable/disable resources),
`pi update --extensions`, and `pi remove <source>`. Inside a session, `/jev status`
shows whether the extension is active. The hosted key is read from the environment at
load time: `export TYPESAFE_API_KEY=...`. `TYPESAFE_BASE_URL` points the same requests at a
self-hosted Jev-compatible server instead, with `TYPESAFE_REQUEST_EXTENSIONS` and
`TYPESAFE_CONFIDENCE` for tuning it (