Skip to main content
FireConnect can route OpenCode, Codex, and Pi through Fireworks on Microsoft Foundry instead of the Fireworks gateway. Usage is billed through Azure and counts toward your Microsoft Azure Consumption Commitment (MACC) where applicable.

Microsoft Foundry portal setup

Enable Fireworks on Foundry, create a deployment, and find your project endpoint — start here if you have not set up a Foundry resource yet
CLI terminology: The Foundry provider is --provider azure (or on --azure). Harness configs display the label Fireworks on Microsoft Foundry. Model IDs are your deployment names (for example, FW-GLM-5.1) — not Fireworks serverless short IDs like glm-fast-latest.

​
Supported harnesses

HarnessAzure in FireConnectNotes
OpenCodeYesfireworks-azure provider in opencode.json
CodexYesfireworks-azure block in config.toml
PiYesfireworks-azure provider in models.json
Claude CodeNot yetclaude on always wires direct Fireworks today
CursorNot yetcursor on always wires direct Fireworks; global --provider azure is ignored
VS CodeNot yetvscode on always wires direct Fireworks today
Deep AgentsNot yetdeepagents on always wires direct Fireworks today
FireConnect implements Azure routing for OpenCode, Codex, and Pi today. The other harnesses still route through the Fireworks gateway only — that is a FireConnect implementation gap, not a Foundry platform limitation. If you run fireconnect configure --provider azure and then fireconnect cursor on (or claude on, etc.), the harness enables direct Fireworks and does not read the global Azure settings.

​
Prerequisites

  • A Microsoft Foundry resource with at least one Fireworks model deployment (for example, FW-GLM-5.1 or FW-MiniMax-M2.5)
  • Your Foundry resource endpoint and Azure API key from the Microsoft Foundry portal
  • OpenCode, Codex, or Pi installed locally
  • FireConnect installed (see Overview — Install)
Use an Azure API key from Foundry — not a Fireworks key (fw_...) or Fire Pass key (fpk_...). Fire Pass is not supported on the Foundry path.

​
Configure once, then enable harnesses

fireconnect configure sets the Foundry provider and endpoint. It does not set your Fireworks API key — use fireconnect login for that when routing through the Fireworks gateway. 
export AZURE_API_KEY=<your-azure-api-key>

fireconnect configure \
  --provider azure \
  --base-url https://<resource>.services.ai.azure.com \
  --api-key $AZURE_API_KEY
FireConnect stores a top-level provider and azure block in ~/.fireconnect/config.json. After configuring, enable harnesses normally: 
fireconnect opencode on --main FW-GLM-5.1
fireconnect codex on --main FW-GLM-5.1
fireconnect pi on --main FW-GLM-5.1
If you omit --main, FireConnect defaults to FW-GLM-5.1.

​
Endpoint normalization

Pass your Foundry endpoint to --base-url. FireConnect normalizes whatever you paste to the correct OpenAI-compatible base at https://<resource>.services.ai.azure.com/openai/v1:
  • Bare resource root (https://<resource>.services.ai.azure.com)
  • Portal project endpoint (.../api/projects/<name>)
  • Foundry Models route (.../models)
  • An already-correct base (.../openai/v1)
Find the endpoint in the Microsoft Foundry portal under Project settings.

​
API key storage

  • Pass --api-key to write the Azure key into ~/.fireconnect/config.json
  • Or export AZURE_API_KEY before configure — FireConnect stores an environment reference instead

​
One-off Foundry routing

Route a single harness through Foundry without changing global config: 
fireconnect opencode on \
  --azure \
  --base-url https://<resource>.services.ai.azure.com \
  --api-key $AZURE_API_KEY \
  --main FW-MiniMax-M2.5
If global config already has a Foundry endpoint, --azure alone reuses it: 
fireconnect opencode on --azure --main FW-GLM-5.1

​
What each harness writes

Each harness writes a dedicated fireworks-azure provider distinct from the Fireworks gateway. off restores your original config byte-for-byte.
HarnessConfig fileProvider IDNotes
OpenCode~/.config/opencode/opencode.jsonfireworks-azure/<deployment>@ai-sdk/openai-compatible adapter; options.baseURL + options.apiKey
Codex~/.codex/config.tomlfireworks-azurewire_api = "chat"; bearer or env_key = "AZURE_API_KEY"
Pi~/.pi/agent/models.json + settings.jsonfireworks-azureopenai-completions provider; key as literal or $AZURE_API_KEY in auth.json
fireconnect <harness> status reports azure as the provider along with the endpoint and deployment name.
fireconnect <harness> model list and model select browse the Fireworks serverless catalog only. They do not list Foundry deployments. With --provider azure, set your deployment with --main on on.

​
Turn off Foundry routing

There are two ways to stop using Microsoft Foundry, depending on what you want next.

​
Switch back to direct Fireworks

Use this when you want to keep FireConnect enabled but route through the Fireworks gateway again instead of your Foundry deployment.
While ~/.fireconnect/config.json has provider: azure, running fireconnect <harness> on without --azure still routes through Foundry. Change the global provider first.
fireconnect configure --provider fireworks

fireconnect login                    # if you have not signed in yet
fireconnect opencode on
fireconnect codex on
fireconnect pi on
Re-running on replaces the fireworks-azure provider with the normal Fireworks gateway config. You do not need to run off first. The Azure endpoint and key remain stored in ~/.fireconnect/config.json but are unused while provider is fireworks. They are used again if you run configure --provider azure later.

​
Remove FireConnect from a harness entirely

Use off to restore the config snapshot from before FireConnect was first enabled for that harness — your original provider settings, byte-for-byte: 
fireconnect pi off
fireconnect opencode off
fireconnect codex off
off removes the fireworks-azure provider from harness config files. It does not change the global provider field in ~/.fireconnect/config.json. If provider is still azure, the next on will route through Foundry again unless you run configure --provider fireworks first. Restart the harness after off if it is already running.

​
Remove FireConnect everywhere

fireconnect uninstall
Disables all harnesses, restores every backup, and removes the CLI.
GoalCommands
Stop Foundry, keep FireConnect on Fireworksfireconnect configure --provider fireworks then fireconnect <harness> on
Undo FireConnect for one harnessfireconnect <harness> off
Undo FireConnect on all harnessesfireconnect uninstall

​
Verify routing

fireconnect opencode status   # provider=azure, base URL, deployment name
fireconnect codex status
fireconnect pi status

​
Per-harness guides

​
Source

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