Skip to main content
FireConnect routes Pi through Fireworks AI models. See the FireConnect overview for install and CLI basics.

​
Prerequisites

​
Enable Fireworks routing

Restart Pi after enabling if it is already running.

​
Using Fire Pass

Use your fpk_... key during login or with --api-key:
FireConnect detects Fire Pass keys and defaults Pi to glm-fast-latest.

​
Default model

Pi routes a single default model. The default is glm-fast-latest.

​
What gets written

FireConnect:
  • Sets defaultProvider / defaultModel in ~/.pi/agent/settings.json
  • Stores the API key in ~/.pi/agent/auth.json ($FIREWORKS_API_KEY when from env, literal with --api-key; never a literal in keychain mode)
  • Writes auth.json at mode 0600
FireConnect snapshots both files under ~/.fireconnect/pi/ before the first change. Running fireconnect pi off restores them byte-for-byte.

​
API key handling

  • If the key comes from the OS keychain, it is written as $FIREWORKS_API_KEY so the secret stays out of settings
  • Keychain mode installs a shell profile hook so Pi can read FIREWORKS_API_KEY at launch
  • Passing --api-key writes the literal key to auth.json instead

​
Browsing and picking models

Fire Pass keys (fpk_...) show Fire Pass-supported routers only.

​
FireRouter mode

fireconnect pi on --router overrides Pi’s built-in Anthropic provider so its existing Claude model catalog is sent through https://router.fireworks.ai/v1. The Fireworks key is supplied through X-FireRouter-Fireworks-Key; an Anthropic API key is also required:
You can switch among Anthropic models inside Pi without re-running FireConnect. FireConnect backs up settings.json, auth.json, and models.json, and fireconnect pi off restores all three byte-for-byte. model select and model reset do not apply in router mode.
In router mode, Pi’s /model picker still shows Anthropic’s built-in model names (for example, Claude Sonnet 4.5). The selected name is passed through FireRouter, but the display label is Pi’s own — it does not reflect the Fireworks model actually serving the request behind the router. Use fireconnect pi status to see the underlying routing.

​
CLI reference

Run fireconnect pi help for all options.

​
Switch models

​
Turn off Fireworks routing

This restores your previous settings.json and auth.json from the backup in ~/.fireconnect/pi/.

​
Use a non-default settings file

​
Fireworks on Microsoft Foundry

Pi supports Fireworks on Microsoft Foundry (CLI: --provider azure or on --azure). See the FireConnect overview and Microsoft Foundry integration guide for portal setup.

​
Configure and enable

One-off routing:

​
What gets written

FireConnect registers a fireworks-azure openai-completions provider in Pi’s models.json (labeled Fireworks on Microsoft Foundry) and sets defaultProvider / defaultModel in settings.json. The Azure API key is stored in auth.json as $AZURE_API_KEY when from the environment, or literally with --api-key. Pass your Foundry deployment name with --main. model list and model select browse the Fireworks serverless catalog only.

​
Turn off Foundry routing

To switch back to the Fireworks gateway:
To remove FireConnect entirely and restore your original settings.json, auth.json, and models.json:
Restart Pi after switching or running off. See Turn off Foundry routing for details on global config behavior and uninstall.

​
Source

FireConnect is open source: github.com/fw-ai/fireconnect