LiteLLM

Add FireRouter to an existing LiteLLM Proxy deployment so developers request one model instead of choosing between open and closed-source models on every call. FireRouter uses bring-your-own-key (BYOK): your Fireworks API key pays for redirected calls, and an Anthropic API key pays for pass-through to Claude Opus 4.8. See Authentication. FireRouter does not store provider keys.

Prerequisites

A running LiteLLM Proxy configured for Fireworks (see LiteLLM integration)
A Fireworks API key (fw_...)
An Anthropic API key (sk-ant-...) for pass-through

Add FireRouter to `config.yaml`

model_list:
  - model_name: accounts/fireworks/routers/firerouter
    litellm_params:
      model: fireworks_ai/accounts/fireworks/routers/firerouter
      api_key: os.environ/FIREWORKS_AI_API_KEY
      extra_headers:
        x-anthropic-api-key: os.environ/ANTHROPIC_API_KEY

Always use the full router resource ID in litellm_params.model. LiteLLM rewrites short names like fireworks_ai/firerouter to accounts/fireworks/models/firerouter, which FireRouter does not recognize. Set ANTHROPIC_API_KEY on the LiteLLM server so pass-through requests include x-anthropic-api-key automatically. Developers do not need to send the Anthropic key on each request when it is configured here.

A server-side Anthropic key is shared across all callers of that proxy deployment. If each developer should bring their own Anthropic key, omit extra_headers and have clients send x-anthropic-api-key on each request instead (see below).

Start or restart the proxy with both keys exported:

export FIREWORKS_AI_API_KEY="fw_..."
export ANTHROPIC_API_KEY="sk-ant-..."
litellm --config config.yaml

Call FireRouter

Clients authenticate to LiteLLM with a virtual key:

curl http://localhost:4000/chat/completions \
  -H "Authorization: Bearer $LITELLM_VIRTUAL_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "accounts/fireworks/routers/firerouter",
    "messages": [{"role": "user", "content": "Say pong in one word."}]
  }'

Optional routing preference:

-H "x-routing-preference: 4"

See Routing preferences for values 1–5.

Per-developer Anthropic keys

If each caller should use their own Anthropic key, omit extra_headers from the model config and have clients send the header on every request:

curl http://localhost:4000/chat/completions \
  -H "Authorization: Bearer $LITELLM_VIRTUAL_KEY" \
  -H "x-anthropic-api-key: $ANTHROPIC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "accounts/fireworks/routers/firerouter",
    "messages": [{"role": "user", "content": "Say pong in one word."}]
  }'

Enable client header forwarding for the accounts/fireworks/routers/firerouter model group if your LiteLLM version does not forward x- headers by default.

API key layout

Key	Who holds it	Used for
Fireworks API key (`fw_...`)	LiteLLM server	FireRouter auth and redirected inference
LiteLLM virtual key	Each developer or service	Proxy authentication and spend tracking
Anthropic API key (`sk-ant-...`)	LiteLLM server (`extra_headers`) or each caller (request header)	Claude pass-through billing

LiteLLM integration: configure Fireworks models in LiteLLM Proxy
Quickstart: direct API call examples
Authentication: header reference
Routing preferences: tune cost vs. quality

Get Started

Serverless

Deployments

Models & Inference

Training

Fire Pass

Fireworks for Work

FireRouter

FireConnect

Administration

Security & Compliance

Integrations

Reference

Prerequisites

Add FireRouter to `config.yaml`

Call FireRouter

Per-developer Anthropic keys

API key layout

​Prerequisites

​Add FireRouter to config.yaml

​Call FireRouter

​Per-developer Anthropic keys

​API key layout

​Related

Prerequisites

Add FireRouter to `config.yaml`

Call FireRouter

Per-developer Anthropic keys

API key layout

Related