Connect Vast.ai
The Vast.ai integration imports your account's GPU consumption into Cloud Intelligence™. Use Cloud Analytics to analyze Vast.ai spend by workload, GPU model, component and day, next to your other clouds.
Prerequisites
- A Vast.ai account with rented instances or serverless endpoints.
- A DoiT account with the Billing Profiles Admin permission.
Connect each Vast.ai account separately.
Create an API key
- Sign in to the Vast.ai console and open Account.
- Go to API Keys and create a key.
- Give it read access to billing and instances. DoiT never writes to your account, so no write permissions are needed.
- Copy the key. Vast.ai shows it only once.
Connect your account
- Sign in to Cloud Intelligence.
- Open Data ingestion and integrations > Integrations and find Vast.ai.
- Select Connect +.
- Enter an Account name to identify the connection in DoiT, and the API key.
- Select Test connection, then Save after the test succeeds.
The connection test reads the current day's charges. A new account with no usage yet can pass the test.
What is imported
| Reporting dimension | Vast.ai value |
|---|---|
| Provider | Vast.ai |
| Billing account | The account name you entered |
| Workload | The serverless endpoint, or the rented instance |
| Service | Serverless Inference, or GPU Instance |
| SKU | The billed component, with the GPU model on GPU time |
| Region | Country of the host machine |
| Date | Usage day, in UTC |
Vast.ai bills four components per instance per day, and each becomes its own line:
| Component | What it covers |
|---|---|
| GPU time | Hours of GPU rental |
| Storage | Hours of instance disk |
| Bandwidth (download) | Inbound data transfer, in GB |
| Bandwidth (upload) | Outbound data transfer, in GB |
Labels include Charge type, Component, Instance, Serverless endpoint, GPU model, Location and the instance template name.
AI spend
All Vast.ai spend is reported as GenAI spend, whether it comes from a serverless endpoint or an instance you rent directly, and appears on the GenAI dashboards. Use the Charge type label to separate serverless from rented instances.
Vast.ai's billing data contains no model name and no token counts, so Vast.ai spend is not broken down by model or by input and output tokens. Reports grouped by model show it without one.
Credits
Vast.ai credit top-ups are not imported. Credit is a prepaid balance rather than consumption, so counting a top-up as cost would double-count the spend it later pays for.
Hardware and location
GPU model, host machine and location come from Vast.ai's list of your current instances, which does not include instances that have already been destroyed. An instance created and destroyed within a few hours is still billed in full, but may be reported without a GPU model or location.
Refresh
DoiT reads your Vast.ai charges every six hours and reloads the last four days each time. Vast.ai bills against the UTC day and charges accrue through the day while your instances run, so the current day's cost rises until that day closes, and a day's figure can still change shortly after midnight UTC.
Troubleshooting
The connection test fails. Check that the key was copied in full and has not been deleted in the Vast.ai console. Vast.ai rejects a missing or invalid key with an authorization error.
No costs appear for a day with usage. Costs are reported against the UTC day. If your instances ran late in your local evening, the spend may fall on the next UTC day.
A cost line has no GPU model. The instance was destroyed before DoiT last read your instance list. See Hardware and location.