@h00w/pi-provider-aihubmix

Pi provider extension that auto-registers AIHubMix models for the pi coding agent.

• Pulls the live model catalog from AIHubMix
• Registers chat/agent models under the aihubmix provider
• Routes Claude → Anthropic Messages, Gemini → Google Generative AI, everything else → OpenAI Chat Completions
• Caches the catalog on disk and refreshes on demand
• Sends an APP-Code header (defaults to DBPV7867) for developer discounts

AI services powered by AIHubMix.

Install

pi install npm:@h00w/pi-provider-aihubmix

Or for local development:

pi install ./path/to/pi-provider-aihubmix
# or one-shot
pi -e ./path/to/pi-provider-aihubmix

Setup

1. Create an API key at console.aihubmix.com (sk-...).
2. Provide the key in either way:

export AIHUBMIX_API_KEY=sk-...

or inside pi:

/login aihubmix

3. Pick a model:

/model auto
/model aihubmix/gpt-5.6-sol
/model aihubmix/claude-sonnet-5

/model, the footer, and --list-models primarily show model ids. Use provider/id when you need to disambiguate.

Models only appear in /model and --list-models after an API key is available (env var, /login aihubmix, or --api-key).

If you install the package but have not configured a key yet, /model no longer errors with Could not refresh aihubmix — log in or export the env var to use models.

Behind an HTTP proxy, this extension reads HTTP(S)_PROXY / NO_PROXY for catalog refresh. Alternatively set NODE_USE_ENV_PROXY=1 so Node's global fetch uses the proxy.

What gets registered

By default the extension:

Rule	Behavior
Type	Only llm / chat models from the catalog
Tools	Keeps models that advertise tools or function_calling
Auto router	Always includes auto (smart router)
Pure chat	Excluded when requireTools is true

Protocol routing:

Model	API	Base URL
claude*	anthropic-messages	https://aihubmix.com
cc-*, anthropic-* (configurable)	anthropic-messages	https://aihubmix.com
gemini* / gemma*	google-generative-ai	https://aihubmix.com/gemini/v1beta
everything else	openai-completions	https://aihubmix.com/v1

cc-* models are AIHubMix Claude Code endpoints and expect the Anthropic Messages wire format. Prefix routing is controlled by forceAnthropicPrefixes.

Configuration

Optional file: ~/.pi/agent/aihubmix.json

{
  "apiKeyEnv": "AIHUBMIX_API_KEY",
  "baseUrl": "https://aihubmix.com/v1",
  "anthropicBaseUrl": "https://aihubmix.com",
  "geminiBaseUrl": "https://aihubmix.com/gemini/v1beta",
  "appCode": "DBPV7867",
  "includeAuto": true,
  "requireTools": true,
  "allow": [],
  "deny": [],
  "protocols": {
    "openai": true,
    "anthropic": true,
    "gemini": true
  },
  "forceAnthropicPrefixes": ["cc-", "anthropic-"],
  "cacheTtlHours": 12
}

Fields

Field	Default	Description
apiKeyEnv	AIHUBMIX_API_KEY	Env var holding the API key
baseUrl	https://aihubmix.com/v1	OpenAI-compatible endpoint
anthropicBaseUrl	https://aihubmix.com	Anthropic Messages endpoint
geminiBaseUrl	https://aihubmix.com/gemini/v1beta	Gemini native endpoint (must include /v1beta for pi)
appCode	DBPV7867	Value for the APP-Code header. Set "" to disable
includeAuto	true	Keep smart-router auto models
requireTools	true	Drop models without tool/function calling
allow	[]	Substring allowlist (empty = all)
deny	[]	Substring denylist
protocols.*	all true	Enable/disable wire protocols
forceAnthropicPrefixes	["cc-", "anthropic-"]	Model id prefixes forced onto Anthropic Messages (startsWith, case-insensitive). Use [] to disable
cacheTtlHours	12	Disk cache lifetime

APP-Code

Requests include:

APP-Code: DBPV7867

Override or disable in config:

{ "appCode": "your-code" }

{ "appCode": "" }

See AIHubMix APP-Code docs.

Commands

Command	Description
/aihubmix-refresh	Force-refresh the model catalog (bypass cache)
/aihubmix-status	Show registered model count, source, and config summary

Cache

Catalog cache file:

~/.pi/agent/aihubmix-models-cache.json

• Fresh cache is reused within cacheTtlHours (including native /model soft refresh)
• Network failures fall back to a stale cache when available
• Offline / allowNetwork=false refresh uses the disk cache only
• Offline first-run falls back to a single auto model
• Catalog HTTP goes through HTTP(S)_PROXY when set (see Setup note above)

Conflict with models.json

If ~/.pi/agent/models.json already defines providers.aihubmix, pi layers that file over this extension. You will see a warning on load. Prefer either:

• let this extension manage the catalog, or
• use modelOverrides in models.json for small tweaks

Development

npm install
npm test
npm run typecheck

# try locally
pi -e .

License

MIT