Overview
Fireworks is an OpenAI-compatible provider in Bifrost with native support for:- Chat Completions via
/v1/chat/completions - Responses API via
/v1/responses - Text Completions via
/v1/completions - Embeddings via
/v1/embeddings - Streaming for chat, responses, and completions
- Tool calling for chat and responses
- Optional Anthropic-compatible mode via
/v1/messages, enabled per key or per alias withuse_anthropic_endpoints
Supported Operations
By default, Fireworks Responses support is native in Bifrost. Requests are sent to Fireworks’
/v1/responses endpoint directly, so fields such as previous_response_id, max_tool_calls, and store are preserved. With use_anthropic_endpoints on, Responses are converted to the Anthropic Messages format instead, and those Responses-only fields do not apply.Setup & Configuration
Configure Fireworks as a provider.- Web UI
- config.json
- API
- Go SDK

- Navigate to Models > Model Providers. Look for Fireworks under Configured Providers. If it is missing, click on Add New Provider and select Fireworks.
- Click Add Key or edit an existing key.
- Set a name for your key.
- Paste your API key directly or use an environment variable (for example,
env.FIREWORKS_API_KEY). - Set Allowed Models to All Models (default) or the specific model allowlist you want this key to serve.
- Save the provider configuration.
Disable Store (optional)
Fireworks stores Responses API conversations by default (store=true) and keeps them for 30 days, even when your account has zero data retention for other endpoints. Clients on the Anthropic SDK or Chat Completions cannot send store, so enable Disable Store to have Bifrost force store=false on every Fireworks Responses API request, overriding any store value the client sends.
- Web UI
- config.json
Open the Fireworks provider, go to the OpenAI Config tab, turn on Disable Store, and click Save Configuration.
/v1/responses endpoint, which is where Bifrost routes /v1/responses traffic and Anthropic SDK traffic that is not using use_anthropic_endpoints. Chat Completions and the Anthropic-compatible /v1/messages endpoint have no store parameter and are unaffected.
Anthropic-Compatible Endpoints (optional)
Fireworks exposes an Anthropic-compatible Messages endpoint (/v1/messages) alongside its default OpenAI-compatible APIs. Setting use_anthropic_endpoints routes Chat Completions and the Responses API through that endpoint instead. Text Completions and Embeddings are unaffected and always use their OpenAI-compatible endpoints.
Authentication does not change between the two modes: Bifrost sends Authorization: Bearer <key> either way.
The setting can be configured per key, and overridden per model alias:
- Key-level - Sets the default endpoint mode for every request made with that key.
- Alias-level - Overrides the key-level default for a single alias, so one key can serve some aliases through the OpenAI-compatible endpoints and others through the Anthropic-compatible endpoint.
- Web UI
- API
- config.json
On the key form, toggle Use Anthropic Endpoints (off by default). To override this for a specific alias, open that alias’s expanded row in the deployments table and toggle Use Anthropic endpoints under Fireworks overrides, which takes priority over the key-level setting for that alias only.
1. Chat Completions
Fireworks chat completions use the standard OpenAI-compatible wire format.Fireworks-specific handling
predictionis preserved and forwarded.- Bifrost maps
prompt_cache_keyto Fireworksprompt_cache_isolation_keyfor chat-completion cache isolation. - Assistant
reasoning_contentis preserved for Fireworks chat-completion models that support reasoning history.
Filtered Parameters
For Fireworks chat completions, Bifrost removes or rewrites a small set of OpenAI-specific fields before sending the request upstream:prompt_cache_keyis mapped to Fireworksprompt_cache_isolation_keyprompt_cache_retentionis removedverbosityis removedstoreis removedweb_search_optionsis removed
Example
2. Responses API
Fireworks Responses use the native Fireworks endpoint:previous_response_idmax_tool_callsstore- native responses streaming
use_anthropic_endpoints enabled on the key or alias, Responses are converted to the Anthropic Messages format and sent to /v1/messages instead. See Anthropic-Compatible Endpoints.
Example
previous_response_id.
3. Text Completions
Fireworks text completions are sent to the native completions endpoint:Example
prompt_cache_key from extra_params and maps it to Fireworks prompt_cache_isolation_key.
4. Embeddings
Fireworks embeddings are sent to:Example
prompt_template, return_logits, and normalize. This page describes the standard embeddings flow currently covered by Bifrost.
5. Unsupported Features
The following operations are still unsupported by the Fireworks provider in Bifrost:6. Caveats
Prompt Caching Semantics
Prompt Caching Semantics
For Fireworks chat completions, Bifrost maps
prompt_cache_key to Fireworks prompt_cache_isolation_key, which is the Fireworks body field for cache isolation. Fireworks also accepts the header form x-prompt-cache-isolation-key. For text completions, Bifrost extracts prompt_cache_key from extra_params and maps it to the same Fireworks body field. If you need Fireworks session-affinity behavior, pass user, configure x-session-affinity in provider extra headers, or send it through the HTTP gateway via x-bf-eh-x-session-affinity. Live cache-hit behavior remains model and deployment dependent.Reasoning History
Reasoning History
Bifrost preserves assistant
reasoning_content for Fireworks chat models that support reasoning history. Fireworks-specific reasoning controls such as reasoning_history are not given special typed handling in this provider page.
