Today, Vantage is announcing support for Fireworks AI costs, giving customers the ability to track Fireworks AI inference, deployment, and training usage and costs directly in Vantage. Once connected, Vantage automatically ingests Fireworks AI's usage and cost data, allowing customers to monitor consumption, set budgets, detect anomalies, and allocate Fireworks AI costs alongside their broader cloud and AI infrastructure.

Fireworks AI is a generative AI platform teams use to run, fine-tune, and deploy open-source and custom models, from serverless inference to dedicated GPU deployments, without managing the underlying infrastructure. Enterprises typically run Fireworks AI alongside major cloud providers or frontier LLM providers, making it a growing part of their overall AI infrastructure spend. Previously, Vantage customers relied on manually intensive Custom Provider uploads of Fireworks AI data that stitched together Fireworks console exports, API pulls, or invoices to approximate usage and cost. This slowed down chargebacks, obscured which models and deployments were driving spend, and made it difficult to catch spend spikes in time.

Now, with the launch of Fireworks AI cost support in Vantage, customers can grant Vantage access to their Fireworks AI account by creating an API key in the Fireworks AI dashboard. Vantage retrieves daily cost and usage data broken down across Fireworks AI's products—including Serverless Inference, Serverless Training API, On-Demand Deployments, and Managed Training—with per-line detail on the model bucket, accelerator type, or base model driving each charge. For example, customers can see how much spend is attributable to cached versus uncached input tokens on serverless inference, track GPU-hours consumed by on-demand deployments by accelerator type, and separate fine-tuning training spend from production inference.

The Fireworks AI integration is now available to all Vantage customers. To connect, go to the Integrations page in account settings and select Fireworks. You can learn more via the Fireworks section in Vantage's documentation.

Frequently Asked Questions

1. What is being launched today?

Vantage is launching native support for Fireworks AI cost and usage data. Customers can now connect Fireworks AI via an API key, enabling automatic ingestion and visibility of Fireworks AI inference, deployment, and training usage and costs in Vantage.

2. Who has access to this integration?

This integration is available to all Vantage customers across all subscription tiers.

3. How much does this integration cost?

There is no additional fee for using the integration. However, Fireworks AI costs will be included in quota tier enforcement. If your Fireworks AI costs push you over your current tier limit, you may be prompted to upgrade. To see more details on pricing, please refer to the Pricing page.

4. How does the integration work?

Vantage uses a Fireworks AI API key to ingest cost data through Fireworks AI's billing summary API at daily granularity. After authorizing Vantage access to your Fireworks AI account, Vantage ingests rated cost data broken down by Fireworks AI product (for example, Serverless Inference or On-Demand Deployments), the model bucket, accelerator type, or base model driving each charge, and the specific usage category, such as uncached LLM input tokens, LLM output tokens, or on-demand deployment GPU-hours.

5. What permissions are required on the Fireworks AI side?

You need to generate a Fireworks AI API key for the account whose billing data you want to track. It's best practice to use a dedicated key for Vantage and rotate it per your security policy.

6. What permissions are needed within Vantage?

You must have a Vantage Owner or Integration Owner role to add or remove the Fireworks AI integration.

7. What permissions does Vantage have in my Fireworks AI account?

Vantage can't perform any actions that incur costs. Vantage only reads billing and usage data from Fireworks AI's billing summary endpoint and will never perform any other actions.

8. What dimensions can Fireworks AI costs be filtered and grouped by?

Fireworks AI reports can support aggregating and filtering on the following dimensions:

• Service (Serverless Inference, Serverless Training API, On-Demand Deployments, Managed Training)
• Category (the model bucket, accelerator type, or base model name driving the charge)
• Subcategory (e.g., cached and uncached LLM input tokens, LLM output tokens, On-demand deployment GPU-hours, Serverless Training Tokens, and Supervised Fine-tuning Tokens)
• Charge Type (Usage)
• Turbo fine-tuning, available as a tag fireworks:is_turbo on Managed Training costs

9. Can I view Usage data for Fireworks AI?

Yes. Usage data is available for inference and training, measured in tokens, and for on-demand deployments, measured in GPU-hours.

10. Are there Active Resources for Fireworks AI?

No, there are no Active Resources to monitor for Fireworks AI.

11. How often does Fireworks AI data refresh in the Vantage console?

Fireworks AI data refreshes daily in the Vantage console.

12. What happens if I remove a Fireworks AI integration?

If you decide to remove your Fireworks AI integration from Vantage, all costs associated with that Fireworks AI account will be removed from the Vantage console.

13. How far back will the data go?

The integration will fetch historical cost data available from Fireworks AI's billing summary API for your account.

14. Can I have multiple Fireworks AI account integrations?

Yes. You can connect multiple Fireworks AI accounts, each with its own API key.

15. My Fireworks AI invoice includes credits and discounts. Will Vantage's costs match my invoice exactly?

Vantage shows gross rated usage cost from Fireworks AI's billing summary. Credits, discounts, taxes, and invoice adjustments are not currently exposed by Fireworks AI's billing summary API, so costs in Vantage reflect gross rated usage and may differ from your final Fireworks AI invoice after credits and adjustments are applied. If you have negotiated pricing you'd like reflected, contact support@vantage.sh.

16. Does Vantage store prompt or response content?

No. Vantage only stores billing metadata (e.g., token and GPU-hour consumption). No user content (prompts or responses) is ingested.