Skip to main content
givemeanode (now Autoresearch by San Francisco Compute; givemeanode.com redirects there) runs small isolated VMs called sandboxes that execute shell commands for agents, alongside GPU and CPU nodes. The following table shows what CostGraph runs on this provider. givemeanode now redirects to Autoresearch by San Francisco Compute. CI runners: no. AutoPilot compares runners by price, and this provider publishes no sandbox price.

​
Connect

Run the dispatcher with provider set to givemeanode and these variables:
  • GMN_TOKEN

​
Key and permissions

GMN_TOKEN is an org service token (gmnt_…). An org admin mints it on the /team page or with autor token create --name <n> --workspace <ws>; it is shown once. It expires in 90 days unless --expires-in-days is passed (0 = never). Documented: a token carries only the access granted. The default access level ‘infra’ covers nodes, sandboxes, exec and jobs, so it is the minimum for create/exec/list/delete. A read-only token can only list and read. The token is limited to the workspaces it is minted for, so mint it for one workspace. Only an org admin can mint tokens. Sources: autoresearch.sfcompute.com: docs#ci and automation, autoresearch.sfcompute.com: docs#sandboxes, autoresearch.sfcompute.com: docs#sandbox limits, autoresearch.sfcompute.com: docs#sandbox sizes, autoresearch.sfcompute.com: docs#roles.

​
Limits

  • Sandbox HTTP API is under the /preview path (“every /preview API call then runs as that token”). (source)
  • Command deadline per exec: 60 seconds by default, 10 minutes at most. Output per command is capped at 1 MiB stdout and 1 MiB stderr. A sandbox runs commands one at a time. (source)
  • Up to 4,096 sandboxes at once per workspace. RAM per sandbox is 8 GiB, raisable on request up to 64 GiB. (source)
  • Sizes: sandbox-sm 1 vCPU/2 GiB (default), sandbox-md 4/8 GiB, sandbox-lg 8/32 GiB, sandbox-xl 16/64 GiB. The sandbox_vcpus limit starts at 4, so lg and xl need a raise. (source)
  • No maximum lifetime is documented. A sandbox bills for its RAM until it is deleted. Idle sandboxes park themselves after 1.5 seconds without a command, giving back memory they do not use, and keep their state. (source)
  • givemeanode runs one API endpoint per region (for example api.use1.givemeanode.com), and endpoints are not interchangeable. The docs do not state a sandbox CPU architecture. (source)