Fireworks AI
Run Fireworks AI chat, embedding, and image models from Swarmz server workflows
Connect Fireworks AI when a project needs an open-model chat completion, text embeddings, or FLUX image generation. Every request runs through the connector gateway, so the Fireworks API key stays out of generated browser code.
The connection is shared by the workspace. Usage and charges belong to the connected Fireworks account.
Availability
A Swarmz administrator must show Fireworks AI under Platform config > Integrations. It does not require a Swarmz OAuth application.
Connect Fireworks AI
- In the Fireworks dashboard, create an API key for the account that owns the intended models. Fireworks documents API key authentication.
- In the Swarmz project, open Settings > Integrations and select Fireworks AI.
- Select Add connection, paste the key into API key, and select Save connection.
Swarmz validates the key against the Fireworks model catalog, encrypts it, and stores it behind the connector gateway.
Available operations
| Operation | What it does | Limit per minute |
|---|---|---|
list_models | Lists up to 200 current serverless models | 20 |
chat_completion | Runs one non-streaming chat with bounded messages and output tokens | 20 |
create_embeddings | Embeds up to 100 bounded text inputs | 20 |
generate_image | Generates one bounded FLUX.1 Schnell image result | 5 |
Inference operations are billable writes and require allow-once approval. Model listing is a read. Image results are returned as bounded JSON and can be stored as a project asset by generated server code.
Ask the AI to use Fireworks AI
- “List suitable Fireworks serverless models, then add a non-streaming support-answer endpoint.”
- “Create embeddings for these help articles and store the vectors with their source IDs.”
- “Generate a 16:9 FLUX image for this event card and save the returned image as an asset.”
The AI lists live models before choosing a text or embedding model. It does not invent a model ID or put the Fireworks key in a client environment variable.
Verify Fireworks AI
- List models and copy one returned model name.
- Run a short chat with a small
max_tokensvalue. - Embed one known sentence and confirm a vector is returned.
- Generate a low-cost test image and check its completion result and seed in Fireworks usage.
Limits and recovery
Chat accepts up to 50 messages and 8,192 output tokens. Embedding accepts up to 100 inputs. Image generation uses FLUX.1 Schnell and supports the aspect ratios shown by the connection operation; other Fireworks image workflows are not exposed by this connector.
If a model request is rejected, list models again and verify that the selected model supports serverless inference for the connected account. To rotate the key, replace it in Swarmz, run list_models, then revoke the old key in Fireworks. Remove the connection to stop all project use.