Microsoft Foundry

FireConnect can route supported harnesses 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. With Foundry, --model is the model you deployed in Azure (for example, FW-GLM-5.2), not a Fireworks serverless short ID like glm-fast-latest.

Supported harnesses

Harness	Azure in FireConnect	Notes
OpenCode	Yes	`fireworks-azure` provider in `opencode.json`
Codex	Yes	`fireworks-azure` block in `config.toml`
Pi	Yes	`fireworks-azure` provider in `models.json`
Cursor	Yes	OpenAI BYOK override pointed at Foundry
VS Code	Yes	Custom chat-completions endpoint in `chatLanguageModels.json`
Deep Agents	Yes	`fireworks-azure` provider in `config.toml`
Claude Code	Not yet	`claude on` always wires direct Fireworks today

Claude Code is the remaining gap. Every other harness in the matrix supports Foundry routing in FireConnect v0.9.0+.

Prerequisites

A Microsoft Foundry resource with at least one Fireworks model deployment (for example, FW-GLM-5.2 or FW-MiniMax-M2.5)
Your Foundry resource endpoint and Azure API key from the Microsoft Foundry portal
A supported harness installed locally
FireConnect v0.9.0+ (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 --model FW-GLM-5.2
fireconnect codex on --model FW-GLM-5.2
fireconnect pi on --model FW-GLM-5.2
fireconnect cursor on --model FW-GLM-5.2
fireconnect vscode on --model FW-GLM-5.2
fireconnect deepagents on --model FW-GLM-5.2

If you omit --model, FireConnect defaults to FW-GLM-5.2.

Cursor and VS Code: fully quit the IDE before on or off. FireConnect writes SQLite state and hard-errors if the app is still running unless you pass --force.

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 \
  --model FW-MiniMax-M2.5

If global config already has a Foundry endpoint, --azure alone reuses it:

fireconnect cursor on --azure --model FW-GLM-5.2

What each harness writes

Each harness writes a dedicated Foundry config distinct from the Fireworks gateway. off restores your original config byte-for-byte.

Harness	Config file	Provider ID	Notes
OpenCode	`~/.config/opencode/opencode.json`	`fireworks-azure/FW-GLM-5.2`	`@ai-sdk/openai-compatible` adapter; `options.baseURL` + `options.apiKey`
Codex	`~/.codex/config.toml`	`fireworks-azure`	`wire_api = "chat"`; bearer or `env_key = "AZURE_API_KEY"`
Pi	`~/.pi/agent/models.json` + `settings.json`	`fireworks-azure`	`openai-completions` provider; key as literal or `$AZURE_API_KEY` in `auth.json`
Cursor	`state.vscdb`	Foundry deployment name	OpenAI BYOK base URL + Azure key; one deployment in the picker
VS Code	`chatLanguageModels.json` + `state.vscdb`	Foundry deployment name	Chat-completions endpoint; Azure key in secret storage
Deep Agents	`~/.deepagents/config.toml`	`fireworks-azure:FW-GLM-5.2`	`openai-completions` provider with Foundry base URL

fireconnect <harness> status reports azure as the provider along with the endpoint and model.

fireconnect model list browses the Fireworks serverless catalog only. It does not list Foundry deployments. With --provider azure, set your model with --model 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
fireconnect cursor on
fireconnect vscode on
fireconnect deepagents on

Re-running on replaces the Foundry config 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 are restored byte-for-byte:

fireconnect pi off
fireconnect opencode off
fireconnect codex off
fireconnect cursor off
fireconnect vscode off
fireconnect deepagents off

off removes Foundry wiring 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. For Cursor and VS Code, quit the IDE before off.

Remove FireConnect everywhere

fireconnect uninstall

Disables all harnesses, restores every backup, and removes the CLI.

Goal	Commands
Stop Foundry, keep FireConnect on Fireworks	`fireconnect configure --provider fireworks` then `fireconnect <harness> on`
Undo FireConnect for one harness	`fireconnect <harness> off`
Undo FireConnect on all harnesses	`fireconnect uninstall`

Verify routing

fireconnect opencode status   # provider=azure, base URL, model
fireconnect codex status
fireconnect pi status
fireconnect cursor status
fireconnect vscode status
fireconnect deepagents status

Per-harness guides

Source

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

Get Started

Serverless

Deployments

Models & Inference

Training

Fire Pass

Fireworks Nexus

FireRouter

FireConnect

Administration

Security & Compliance

Integrations

Reference