pi-9router

9Router AI model provider extension for pi coding agent. 9Router is a local/remote AI gateway that exposes an OpenAI-compatible REST API. The extension dynamically fetches the live model list from the gateway at startup and registers chat-capable models with pi.

Installation

pi install npm:@neilurk12/pi-9router

Prerequisites

Start 9Router locally before launching pi. See the 9Router setup guide for installation. By default, the extension expects 9Router at http://localhost:20128.

Verify 9Router is reachable:

curl http://localhost:20128/api/health
# {"ok":true}

Setup

The gateway URL and API key are both optional — local 9Router installs typically run with auth disabled and on the default port.

Environment variables (recommended)

export NINEROUTER_URL="http://localhost:20128"   # default; override for VPS/tunnel
export NINEROUTER_KEY="sk-..."                  # only if your 9router requires auth

Config file

Create ~/.pi/9router.json:

{
  "baseUrl": "https://9router.example.com",
  "apiKey": "sk-..."
}

If both env vars and the config file are set, environment variables take precedence.

Models

Models are fetched from ${NINEROUTER_URL}/v1/models at every pi startup — no hardcoded list. Whatever you configure in 9Router shows up in pi's /model picker.

Entries with non-chat kind values (e.g. webSearch, embedding, tts, stt, image) are filtered out, so only chat-capable models appear in /model.

Usage

Select a 9Router model in pi with the /model command.

Troubleshooting

• No 9router models in /model — check that 9Router is running (curl $NINEROUTER_URL/api/health) and that NINEROUTER_URL points to it. The extension logs the exact fetch URL on failure.
• 401 Unauthorized — set NINEROUTER_KEY or add apiKey to ~/.pi/9router.json.
• Wrong models listed — reconfigure 9Router; the extension re-fetches on every startup.