Connect Fireworks AI ​

Fireworks AI hosts open-weight models, such as Llama, DeepSeek, Qwen, Kimi, and GLM, behind an OpenAI-compatible API. Connect your Fireworks AI API key to Model Gateway to route requests to the open-weight models Fireworks AI serves.

Model Gateway stores your Fireworks AI API key in a provider connection. Applications authenticate with Model Gateway access keys instead of your Fireworks AI API key, and policies apply rate limits, token limits, and IP restrictions to Fireworks AI traffic the same way they apply to every other provider.

Prerequisites ​

You must have the following to connect Fireworks AI to Model Gateway:

• A Fireworks AI account with available credits. Fireworks AI suspends accounts with no credits, and a suspended account can't create API keys or send requests. Add credits or a payment method in Billing.
• Access to Model Gateway in your Workato workspace

Fireworks AI API details ​

Model Gateway connects to Fireworks AI through the OpenAI Compatible provider type. Use the following values when you configure the connection:

Refer to the Fireworks AI OpenAI compatibility documentation for more information about the Fireworks AI API.

Create a Fireworks AI API key ​

Complete the following steps to create an API key in Fireworks AI:

Add Fireworks AI as a provider ​

Complete the following steps to add Fireworks AI as a provider in Model Gateway:

The provider displays in the Providers list with a Connected status.

Find a Fireworks AI model ID ​

Each route sends requests to one Fireworks AI model, which you specify by its model ID in the route's Model field. The Model field suggests Fireworks AI models, and the model library shows each model's capabilities so you can choose one or find a model that isn't suggested. Fireworks AI identifies each model by its full path, not its display name.

Complete the following steps to find a model ID:

Fireworks AI model support varies by model. For example, only some models support function calling or image input. Check Supporting Functionality on the model's page before you route production traffic to it.

Create a route to a Fireworks AI model ​

A direct route sends every request to the same provider and model. Complete the following steps to create a direct route to a Fireworks AI model:

Refer to Routes to create rule-based and intelligent routes. For example, a rule-based route can send requests from one team to a Fireworks AI model and send all other requests to a different provider. Refer to Route builder to define the rules.

Create an access key ​

Create an access key that includes the Fireworks AI route. Applications use this key to authenticate with Model Gateway instead of your Fireworks AI API key. Refer to Create an access key.

Send a request to a Fireworks AI model ​

Authenticate with your Model Gateway access key and pass the route ID as the model value. Model Gateway resolves the route to your Fireworks AI provider and model, applies any assigned policy, and forwards the request to Fireworks AI with your Fireworks AI API key.

curl <model-gateway-url> \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <access-key>" \
  -d '{
    "model": "<route-id>",
    "messages": [
      {
        "role": "user",
        "content": "Hello"
      }
    ]
  }'

from openai import OpenAI

client = OpenAI(
    base_url="<model-gateway-host>/v1",
    api_key="<access-key>",
)

response = client.chat.completions.create(
    model="<route-id>",
    messages=[{"role": "user", "content": "Hello"}],
)

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

The model field in the response shows the Fireworks AI model that served the request, for example accounts/fireworks/models/gpt-oss-120b.

Model Gateway also accepts Anthropic-compatible requests for routes backed by Fireworks AI. Refer to Test a route for an Anthropic-compatible example, and Connect clients to configure Claude Code and the Anthropic SDKs.

Fireworks AI behavior ​

The following behavior applies to Fireworks AI requests you send through Model Gateway:

• Context length: Fireworks AI reduces max_tokens by default when the prompt plus max_tokens exceeds the model's context window. OpenAI returns an error instead. Set context_length_exceeded_behavior to error in the request body to return an error.
• Function calling: Fireworks AI supports OpenAI-compatible function calling on models that support it. Check the model's capabilities before you rely on function calling or parallel function calls.
• Reasoning models: Reasoning models, such as gpt-oss-120b, return their reasoning in the reasoning_content field of the response message, in addition to the answer in the content field.
• Streaming: Fireworks AI supports streaming responses. Model Gateway supports Server-Sent Events (SSE) end to end, so streaming responses from Fireworks AI render incrementally.

Usage and billing for Fireworks AI models apply to your Fireworks AI account. Refer to the Fireworks AI documentation for model pricing and rate limits.

Troubleshooting ​

Use the following troubleshooting information to fix failed requests to a Fireworks AI route. Error responses that come from Fireworks AI include "error_from_provider": true, and the error.message field contains the Fireworks AI error message. Errors without this field come from Model Gateway, which means the request didn't reach Fireworks AI.

Access key rejected ​

Model Gateway returns 401 and the message can't authorize request: invalid credentials when it doesn't recognize the access key in the Authorization header. This error often occurs when the header contains extra characters, such as a line break or a key pasted more than once. Confirm the header uses the format Bearer <access-key> with the full access key, or create a new access key if you no longer have the original.

Provider unreachable ​

Model Gateway returns Error 503 when it can't reach the provider while you add or edit it. Confirm the API URL field is https://api.fireworks.ai/inference/v1, then click Create or Save again.

Fireworks AI API key rejected ​

Fireworks AI returns the message The API key you provided is invalid. when the Fireworks AI API key on the provider is invalid or expired. Confirm the key is active in Fireworks AI, then edit the provider to enter a new key if needed.

A blank API URL field causes a similar error. Model Gateway sends requests to OpenAI when the API URL field is blank, and OpenAI rejects the Fireworks AI API key with the message Incorrect API key provided. Edit the provider and enter https://api.fireworks.ai/inference/v1 in the API URL field to resolve this error.

Fireworks AI account suspended ​

Fireworks AI suspends accounts that have no credits and returns the message account is suspended due to CREDIT_DEPLETED. Add credits or a payment method in Fireworks AI Billing to reactivate the account.

Model not found ​

Fireworks AI returns the message Model not found, inaccessible, and/or not deployed when the route's Model field doesn't match a model it serves. This error occurs when the field contains a display name instead of a model ID, or when the model doesn't support serverless. Edit the route and confirm the Model field uses the full model path, for example accounts/fireworks/models/gpt-oss-120b. Refer to Find a Fireworks AI model ID to confirm the model supports serverless.

Route not found ​

Model Gateway returns 404 and the message model "<route-id>" not found when it can't find the route ID in the model value of the request. Confirm the model value matches the Route ID on the route. Then confirm the access key includes the route in its Route(s) list, and edit the access key to add the route if needed.