pi-9router

A pi package that connects your agent to 9router and enriches its model catalog with metadata from models.dev.

What it does

• Discovers the models exposed by your local (or remote) 9router instance.
• Looks up each model in the public models.dev API to obtain:
  • context window size
  • max output tokens
  • reasoning / thinking support
  • supported input modalities (text, image)
  • pricing (input / output / cache read / cache write)
• Registers 9router as a pi provider so you can select models with /model.
• Provides /9router-status, /9router-refresh, /9router-set-key and /9router-remove-key commands.

Installation

Install as a pi package:

pi install git:github.com/es11225599/pi-9router

Or clone and install from the local path:

cd /path/to/pi-9router
pi install .

Configuration

9router does not require an API key by default, so models appear in /model without any configuration. If your 9router instance is protected by a key, set it via environment variable:

export NINEROUTER_BASE_URL="http://localhost:20128/v1"   # default
export NINEROUTER_API_KEY="<your 9router dashboard API key>"

The API key is created in the 9router dashboard under Settings → API Keys.

If you prefer not to use an environment variable, you can save the key directly from pi with:

/9router-set-key <your-9router-api-key>

The key is stored in pi's auth.json. Use /9router-remove-key to delete it.

Note: If your 9router does not require a key, leave NINEROUTER_API_KEY unset and do not run /9router-set-key. All discovered models will still be available in /model.

Usage

After installation, 9router models appear in pi automatically. Select one with:

/model 9router/openai/gpt-4o

Then send prompts as usual. pi forwards the request to 9router's OpenAI-compatible endpoint (/v1/chat/completions) using the model id without the provider prefix (e.g. openai/gpt-4o), which is exactly what 9router expects.

Commands

• /9router-status — show the configured base URL, discovered model count, API key status and how many models are available in /model.
• /9router-refresh — re-fetch the model list from 9router and re-register the provider.
• /9router-set-key <key> — save the 9router API key to pi's auth.json.
• /9router-remove-key — remove the saved 9router API key.

How the mapping works

9router models are exposed as alias/modelId, for example:

• openai/gpt-4o
• gemini/gemini-2.5-pro
• kc/anthropic/claude-sonnet-4-20250514 (Kilo connection exposing Anthropic models)
• openrouter/google/gemma-4-26b-a4b-it:free

The extension maps the 9router alias to a models.dev provider id (e.g. gemini → google, kc → kilo), then matches the model id by:

1. exact id match
2. normalized id match (strips date suffixes such as -20250514)
3. model name matching
4. nested provider alias matching (e.g. kc/anthropic/... → anthropic)

Models that cannot be matched fall back to sensible defaults (128k context window, 8k output, no reasoning, zero cost).

Project structure

pi-9router/
├── package.json          # npm + pi manifest
├── src/
│   ├── index.ts          # extension entry point
│   ├── config.ts         # env var / default configuration
│   ├── auth.ts           # pi auth.json persistence helpers
│   ├── ninerouter.ts     # 9router /v1/models client
│   ├── modelsdev.ts      # models.dev API client
│   └── mapping.ts        # alias mapping and model enrichment logic
├── scripts/
│   ├── test-mapping.ts            # unit test for the mapping
│   ├── test-extension-load.ts     # smoke test loading the extension
│   ├── test-auth.ts               # test API key persistence
│   └── test-real-mapping.ts       # live test against a running 9router
└── README.md

Development

npm install
npm run check          # TypeScript type check
npm run test:mapping   # test mapping against cached models.dev data
npm run test:auth      # test API key persistence
npm run test:extension # smoke test loading the extension

To test against the live 9router instance:

node --import jiti/register scripts/test-real-mapping.ts

License

MIT