FireRouter Setup

Use the same model ID, such as auto or firerouter/opus, from a coding harness, an LLM gateway, or your own code. Every path needs a Fireworks API key. Each closed provider in the route also needs a credential. See Provide Anthropic and OpenAI credentials.

You are using	Setup
Claude Code, Codex, OpenCode, Cursor IDE, or another coding harness	FireConnect
LiteLLM, Portkey, or another LLM gateway	LLM gateway
Your own app, agent, or script	API and SDK

Choose a model ID

Model ID	What it routes across	Credentials
`auto`	Fireworks open models only	Fireworks API key
`firerouter`	Claude Opus or GPT Sol, picked from your credentials and client, plus the open-model mix	Fireworks key, plus an Anthropic or OpenAI credential
`firerouter/opus`, `firerouter/sol`, `firerouter/astra`	The named closed family, plus the open-model mix	Fireworks key, plus the matching provider credential
`firerouter/opus/glm-5p3`	Only the models you list	Fireworks key, plus a credential for each closed provider listed

Start with auto if you only have a Fireworks key. Without a closed-model credential, any firerouter ID serves every turn with open models. See Build a router ID and Supported models.

FireConnect

Use FireRouter in any coding harness. FireConnect writes the settings for you and restores them with off. Manual setup gives you the same settings to add yourself, with nothing to install.

Set up your harness

Pick your harness, a model or router ID, and a setup method. Start with auto to route across Fireworks open models with only a Fireworks key, or choose Build your own to compose a firerouter route with Claude, GPT, and open models. The commands and config files update as you choose, including the provider-key headers your route needs.

FireConnect quick start for Claude Code

Install and sign in

curl -fsSL https://fireconnect.fireworks.ai/install.sh | bash
fireconnect login

Connect with a router

fireconnect claude --model firerouter

Restart and pick FireRouter

Start a new session and open /model. Routes with Claude models use your Claude login.

Watch routing live

fireconnect claude live

A meter beside Claude Code shows the model selected for each turn. See Watch routing live.

Which harnesses can send provider keys

Harness	FireConnect with Claude routes	FireConnect with GPT routes	Manual setup
Claude Code	Your Claude login, or `--anthropic-api-key`	OpenAI Provider Key	Sends `x-anthropic-api-key` and `x-openai-api-key`
Codex CLI, Codex app, ChatGPT desktop, OpenCode, Pi, VS Code	`--anthropic-api-key`, or an Anthropic Provider Key	OpenAI Provider Key	Sends `x-anthropic-api-key` and `x-openai-api-key`
Copilot CLI, DeepSeek Harness	Not supported yet	OpenAI Provider Key	Sends `x-anthropic-api-key` and `x-openai-api-key`
Cursor IDE, Copilot App	Not supported	OpenAI Provider Key	Provider Keys only; these apps cannot send extra headers

FireConnect has no local OpenAI key option yet, so GPT routes such as firerouter/astra and firerouter/sol need an OpenAI Provider Key when you connect with FireConnect. To send your own OpenAI key, use manual setup. For every harness’s files and restore behavior, see Coding Harnesses and Harness Compatibility.

LLM gateway

You do not need FireConnect behind a gateway. The gateway sends the exact firerouter/... model ID and your Fireworks API key to Fireworks. Clients and harnesses behind the gateway call the router ID with only their gateway key.

Set up your gateway

Choose LiteLLM or Portkey, pick a preset or build your own route, and choose where the provider keys live. The widget writes the gateway config and a test request. Clients never need provider keys when the gateway stores them or when Provider Keys is connected. For other gateways and more detail on each credential option, see LLM Gateways.

API and SDK

FireRouter works on all three Fireworks inference APIs. Use the SDK you already have and point it at Fireworks.

API	Endpoint	SDKs
Chat Completions (OpenAI-compatible)	`https://api.fireworks.ai/inference/v1/chat/completions`	OpenAI SDK, Fireworks SDK
Responses (OpenAI-compatible)	`https://api.fireworks.ai/inference/v1/responses`	OpenAI SDK
Messages (Anthropic-compatible)	`https://api.fireworks.ai/inference/v1/messages`	Anthropic SDK, Fireworks SDK

In every SDK, the API key is your Fireworks API key. Send closed-provider keys as extra headers:

Header	When
`x-anthropic-api-key`	The route includes a Claude model and no Anthropic Provider Key is connected
`x-openai-api-key`	The route includes a GPT model and no OpenAI Provider Key is connected

Leave out a header when the matching Provider Key is connected. A header, when sent, takes precedence over the stored key for that request. Amazon Bedrock has no request header; it uses Provider Keys only. Install the SDKs you use:

pip install openai anthropic fireworks-ai

npm install openai @anthropic-ai/sdk

Build your request

Pick your models, API, and language. The router ID and the command update as you choose, with the provider headers your route needs.

Examples

The examples below use two router IDs:

firerouter/astra in the Chat Completions and Responses examples. FireRouter routes between GPT Astra and the Fireworks open-model mix, so it needs an OpenAI key. firerouter/sol works the same way with GPT Sol.
firerouter/opus in the Messages examples. FireRouter routes between Claude Opus and the Fireworks open-model mix, so it needs an Anthropic key.

Any router ID works on any API. To use a different route, change model and send the header for each closed provider in it. See Supported models and Build a router ID.

Chat Completions

curl https://api.fireworks.ai/inference/v1/chat/completions \
  -H "Authorization: Bearer $FIREWORKS_API_KEY" \
  -H "x-openai-api-key: $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "firerouter/astra",
    "messages": [{"role": "user", "content": "Say pong in one word."}]
  }'

import os

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["FIREWORKS_API_KEY"],
    base_url="https://api.fireworks.ai/inference/v1",
    # Omit when an OpenAI Provider Key is connected.
    default_headers={"x-openai-api-key": os.environ["OPENAI_API_KEY"]},
)

response = client.chat.completions.create(
    model="firerouter/astra",
    messages=[{"role": "user", "content": "Say pong in one word."}],
)

print(response.model, response.choices[0].message.content)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.FIREWORKS_API_KEY,
  baseURL: "https://api.fireworks.ai/inference/v1",
  // Omit when an OpenAI Provider Key is connected.
  defaultHeaders: { "x-openai-api-key": process.env.OPENAI_API_KEY },
});

const response = await client.chat.completions.create({
  model: "firerouter/astra",
  messages: [{ role: "user", content: "Say pong in one word." }],
});

console.log(response.model, response.choices[0].message.content);

import os

from fireworks import Fireworks

# Reads FIREWORKS_API_KEY from the environment.
client = Fireworks(
    # Omit when an OpenAI Provider Key is connected.
    default_headers={"x-openai-api-key": os.environ["OPENAI_API_KEY"]},
)

response = client.chat.completions.create(
    model="firerouter/astra",
    messages=[{"role": "user", "content": "Say pong in one word."}],
)

print(response.model, response.choices[0].message.content)

Responses

curl https://api.fireworks.ai/inference/v1/responses \
  -H "Authorization: Bearer $FIREWORKS_API_KEY" \
  -H "x-openai-api-key: $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "firerouter/astra",
    "input": "Say pong in one word."
  }'

import os

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["FIREWORKS_API_KEY"],
    base_url="https://api.fireworks.ai/inference/v1",
    # Omit when an OpenAI Provider Key is connected.
    default_headers={"x-openai-api-key": os.environ["OPENAI_API_KEY"]},
)

response = client.responses.create(
    model="firerouter/astra",
    input="Say pong in one word.",
)

print(response.model, response.output_text)

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.FIREWORKS_API_KEY,
  baseURL: "https://api.fireworks.ai/inference/v1",
  // Omit when an OpenAI Provider Key is connected.
  defaultHeaders: { "x-openai-api-key": process.env.OPENAI_API_KEY },
});

const response = await client.responses.create({
  model: "firerouter/astra",
  input: "Say pong in one word.",
});

console.log(response.model, response.output_text);

Messages

curl https://api.fireworks.ai/inference/v1/messages \
  -H "Authorization: Bearer $FIREWORKS_API_KEY" \
  -H "x-anthropic-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "firerouter/opus",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Say pong in one word."}]
  }'

import os

import anthropic

client = anthropic.Anthropic(
    api_key=os.environ["FIREWORKS_API_KEY"],
    base_url="https://api.fireworks.ai/inference",
    # Omit when an Anthropic Provider Key is connected.
    default_headers={"x-anthropic-api-key": os.environ["ANTHROPIC_API_KEY"]},
)

message = client.messages.create(
    model="firerouter/opus",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Say pong in one word."}],
)

text = next(block.text for block in message.content if block.type == "text")
print(message.model, text)

import Anthropic from "@anthropic-ai/sdk";

const client = new Anthropic({
  apiKey: process.env.FIREWORKS_API_KEY,
  baseURL: "https://api.fireworks.ai/inference",
  // Omit when an Anthropic Provider Key is connected.
  defaultHeaders: { "x-anthropic-api-key": process.env.ANTHROPIC_API_KEY },
});

const message = await client.messages.create({
  model: "firerouter/opus",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Say pong in one word." }],
});

const text = message.content.find((block) => block.type === "text");
console.log(message.model, text?.type === "text" ? text.text : "");

import os

from fireworks import Fireworks

# Reads FIREWORKS_API_KEY from the environment.
client = Fireworks(
    # Omit when an Anthropic Provider Key is connected.
    default_headers={"x-anthropic-api-key": os.environ["ANTHROPIC_API_KEY"]},
)

message = client.messages.create(
    model="firerouter/opus",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Say pong in one word."}],
)

text = next(block.text for block in message.content if block.type == "text")
print(message.model, text)

The Anthropic SDK base URL is https://api.fireworks.ai/inference, without /v1. The SDK adds /v1/messages. Some models return a thinking block before the text, so read the first text block rather than content[0].

Check which model served the request

Each example prints the response model field. It names the model that served the turn, such as glm-5p3-flash, claude-opus-5-5, or gpt-6-astra, not the router ID you sent. To confirm closed-model access, add x-routing-preference: 1 so the route’s primary model serves the turn. See Verify routing. For errors, limits, and more API details, see APIs and SDKs.

Nexus

Connect

Choose models

Operate

Reference

Choose a model ID

FireConnect

Set up your harness

Which harnesses can send provider keys

LLM gateway

Set up your gateway

API and SDK

Build your request

Examples

Chat Completions

Responses

Messages

Check which model served the request

​Choose a model ID

​FireConnect

​Set up your harness

​Which harnesses can send provider keys

​LLM gateway

​Set up your gateway

​API and SDK

​Build your request

​Examples

​Chat Completions

​Responses

​Messages

​Check which model served the request

Choose a model ID

FireConnect

Set up your harness

Which harnesses can send provider keys

LLM gateway

Set up your gateway

API and SDK

Build your request

Examples

Chat Completions

Responses

Messages

Check which model served the request