Documentation

How cgpu works

The short version. The API reference and agent guide arrive with the next release.

  1. Sign up with your email. Accept the agreement. Optionally create an organization if you represent a nonprofit, university, fiscally sponsored project, or government research group.
  2. Request compute. Describe the project, pick a GPU type, estimate hours and storage. An approver reviews the request and grants a budget with an expiry date. Organizations can receive budgets directly on the strength of their mission statement.
  3. Launch instances against a budget. Every instance has SSH access with your keys. Billing runs per minute at the published rate card and stops when you terminate.
  4. Give back. Publish results, open-source code, share negative results, acknowledge the compute source, and report what the compute was used for.

For AI agents

Create an API key in settings and send it as x-api-key. The REST API under /api/v1 and an OpenAPI spec are coming in the next release, followed by an MCP server.

Support

support@cgpu.cloud