Skip to main content
Coding agents start at Nexus. Use the Quickstart for harnesses, custom agents, APIs, SDKs, and LLM gateways. Use FireRouter for model routers, the supported model list, composition, closed-model credentials, and pricing. Prefer short model IDs such as firerouter/opus. In LiteLLM litellm_params.model, use fireworks_ai/accounts/fireworks/routers/firerouter/opus. Family names such as opus track the latest evaluated family version. FireConnect is the easiest way to use Fireworks models in the coding harness you already use. It is an open-source CLI that rewrites your harness settings to call Fireworks, adds Fireworks models to the model picker, and restores your original settings when you turn it off. You do not need to host a service or edit config files by hand. Command line

Claude Code

Codex CLI

OpenCode

Pi

Copilot CLI

DeepSeek

Desktop and IDE apps

Claude Desktop

ChatGPT desktop app

Cursor IDE

VS Code

Copilot App

​
Quick start

Install

Requires Node.js 18 or later. On Windows, run it from Git Bash; PowerShell breaks the script’s line endings.

Sign in

Sign in with your browser, or paste a Fireworks key.

Connect a harness

Replace claude with any harness from Coding Harnesses. For Claude Desktop, run fireconnect claude-desktop on. For the ChatGPT desktop app, run fireconnect chatgpt; it shares a config with the Codex CLI, so it connects both. cursor means Cursor IDE only; Cursor CLI is not supported.Not sure of the name? Run fireconnect on its own and pick from the interactive menu. The ChatGPT desktop app is the codex row.

Restart and check

status shows the configured model and the Fireworks models in the picker.
Rather edit settings yourself? Each harness in Coding Harnesses has a Manual setup tab with the exact settings FireConnect would write.

​
Interactive menu

Run fireconnect with no arguments to pick a harness and an action from a menu:
Terminal running fireconnect: the menu lists each harness with its status, codex is selected, then on, and FireConnect connects Codex and prints the direct command, fireconnect codex

Connecting Codex from the interactive menu

  • Each harness shows its status: on, off, off · detected (installed but not connected), or not detected.
  • Use the arrow keys and Enter, or press 1-9 to pick a row directly. Esc goes back one menu, or exits from the top menu.
  • Pick a harness, then on, off, or status. Claude Code also offers usage and live.
  • configure, key, and help sit at the bottom of the list.
  • After the action runs, FireConnect prints the direct command, such as fireconnect codex, so you can skip the menu next time.
FireConnect interactive menu listing claude, claude-desktop, opencode, codex, pi, cursor, vscode, copilot-app, copilot-cli, and deepseek with their status, then configure, key, and help

The FireConnect menu

​
In Claude Code

Run fireconnect claude, then start a new session and open /model. It lists auto, FireRouter, and the Fireworks models next to Claude Code’s own models:
Claude Code model picker with FireRouter selected and Auto, DeepSeek, GLM, Kimi, and MiniMax options

Fireworks models in the Claude Code model picker

Your Claude login stays signed in. See Claude Code for tier pins, the status line, and manual setup.

​
In Codex

Close open Codex sessions, then connect:
Start a new Codex session. /model lists FireRouter, then Codex’s GPT models, then the Fireworks models:
Codex CLI Select Model and Effort picker with FireRouter (current) selected, described as routing each user turn between open models and the closed models you have access to, followed by GPT-6.1-Sol, GPT-6-Astra, GPT-6-Sol, GPT-6-Luna, and older GPT models

The Codex CLI /model picker after fireconnect codex

Codex keeps its own sign-in. Your Fireworks key pays for Fireworks models, and your Codex sign-in pays for GPT models. See Codex for sign-in, billing, and manual setup.

​
In the ChatGPT desktop app

Quit the ChatGPT desktop app, then connect:
Reopen the app. The Fireworks models appear in its model list next to the GPT models. The app shares its config with the Codex CLI, so this also connects Codex. See ChatGPT desktop app.

​
Connect and restore

  • Connect rewrites the harness config to call Fireworks and saves a snapshot under ~/.fireconnect/.
  • off restores file-based harnesses from that snapshot byte-for-byte. For IDE databases, it removes the settings and models FireConnect added.
  • fireconnect upgrade updates FireConnect and brings connected harnesses up to date, keeping the model you picked. Restart each harness the upgrade output names.
  • Sign-in credentials and harness configurations are stored separately. fireconnect login saves your credential through FireConnect’s secret store. If a harness requires the key in its own configuration, FireConnect copies it there and restricts the file permissions. Run fireconnect status to see the active storage backend without revealing the key.

​
Choose a model

Connecting registers coding-ready Fireworks models, including auto. Use --model to add or select a specific model:
Restart or reopen the harness after changing a model. Browse IDs with fireconnect model list. See Open Models for aliases and pinned versions, and Coding Harnesses for app-specific rules. To route each turn between open models and the Claude or GPT models you have access to, use FireRouter. FireRouter Setup covers the FireConnect command and the credentials each harness can send. Web search works through Fireworks with no extra setup. In Claude Code, the native WebSearch and WebFetch tools run server-side on Fireworks. The Codex CLI and the ChatGPT desktop app keep their native web search. Search works with any Fireworks model or FireRouter ID. Web search pricing will be published soon; see Web Search.
Earlier FireConnect versions installed a fireworks-websearch MCP server for Claude Code. That MCP is no longer available. Run fireconnect upgrade to remove it; your own MCP servers are not changed. See Web Search.

​
Upgrade and uninstall

For upgrade and uninstall commands, see the CLI Reference.