Skip to main content
Use LiteLLM Proxy as a shared gateway for Fireworks serverless models, router endpoints, and deployments. LiteLLM exposes a single OpenAI-compatible API to your developers while you manage provider credentials and access control on the server.

Prerequisites

Configure Fireworks models

Create a config.yaml with one entry per model you want to expose. Use the fireworks_ai/ provider prefix and the full Fireworks model or router ID:
Always use full resource paths (for example accounts/fireworks/models/glm-5p2 or accounts/fireworks/routers/firerouter). LiteLLM rewrites short names to accounts/fireworks/models/..., which breaks router endpoints like FireRouter.
LiteLLM sends requests to https://api.fireworks.ai/inference/v1 by default. See the LiteLLM Fireworks AI provider docs for direct-route deployments and other options.

Start the proxy

Call a model

Clients authenticate to LiteLLM with a virtual key:

API key layout

A common pattern is one Fireworks service-account API key on the LiteLLM server, with per-developer virtual keys for access control and attribution.

FireRouter

To add automatic routing between closed-source and open models, register FireRouter in the same model_list. FireRouter requires an Anthropic API key for pass-through — see FireRouter with LiteLLM.