@qmahyar/pi-9router

Connect pi to 9Router — multi-provider chat models plus image, speech, search, and fetch tools

Packages

Package details

extension

Install @qmahyar/pi-9router from npm and Pi will load the resources declared by the package manifest.

npm repo home report

$ pi install npm:@qmahyar/pi-9router

Package: @qmahyar/pi-9router
Version: 1.2.8
Published: Aug 22, 2026
Downloads: 964/mo · 83/wk
Author: qmahyar
License: MIT
Types: extension
Size: 148.7 KB
Dependencies: 0 dependencies · 3 peers

Pi manifest JSON

{
  "extensions": [
    "./extensions/9router.ts",
    "./extensions/9router-tools.ts"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

Install

pi install npm:@qmahyar/pi-9router

Or from git:

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

Requires a running 9Router instance (npm i -g 9router → default http://localhost:20128).

What you get

Command	Role
`/9router`	Connect · full/quick sync · diagnose · register chat models as provider `9router`
`/9router-tools`	Enable tools · set default models · output folder

Tool	On by default	Does
`nr_image_generate`	Yes	Text (or reference images) → image file
`nr_tts`	Yes	Text → speech file
`nr_video_generate`	Yes	Text/image → MP4 (Grok Imagine, async)
`nr_web_search`	Yes	Live web search
`nr_web_fetch`	Yes	URL → markdown
`nr_embed`	No	Text → embeddings
`nr_stt`	No	Audio file → transcript

Off tools leave the model context. Only enabled tools expose schema + usage guidelines to the agent.

Model names

Sync uses smart enrich: when a list row already has name + capabilities (common for chat), it skips /v1/models/info. Thin rows still get a lookup so display names are real — openrouter/openai/tts-1-hd shows as TTS-1 HD, not "Openai/Tts 1 Hd".

Full sync — all tool catalogs + voice TTS probes
Quick sync — chat only (keeps previous image/tts/web catalog)
Diagnose — health, per-kind latency, sample info, voice probes

Tool descriptions stay compact: they name the configured default model, not the whole catalog. A model argument is resolved locally first — nano-banana maps to a real id or fails with candidates listed, instead of No credentials for provider: nano. Browse the full catalog via /9router-tools.

edge-tts / google-tts are free and absent from /v1/models/tts. Full sync probes them and only adds live ones.

60-second start

1. 9router                          # start the gateway
2. /9router  →  Sync models (full catalog)
3. /model    →  provider 9router
4. /9router-tools  →  pick defaults

Optional: Quick sync when you only need chat models refreshed; Diagnose when something is slow or voice TTS is missing.

Pair with Exa (optional)

For dedicated Exa neural search with multi-key rotation (separate from 9Router’s web tools):

pi install npm:@qmahyar/pi-exa-search

→ @qmahyar/pi-exa-search · npm

Use one search stack at a time if you want to avoid overlapping tools.

Docs

Doc
Setup	Install, first run, env vars
Usage	Menus, tools, on/off behavior
Dev	Layout for contributors
Changelog	Unreleased WIP + version history

License

MIT