Skip to main content
Agent37 runs persistent isolated Linux instances (containers or full VMs) that you create from a template and drive over a REST API, including a command exec endpoint. The following table shows what CostGraph runs on this provider.

Connect

Run the dispatcher with provider set to agent37 and these variables:
  • AGENT37_API_KEY
To see what Agent37 costs, connect the Agent37 integration.

Key and permissions

AGENT37_API_KEY is an Agent37 API key (sk_live_…), created in the dashboard at Cloud > API keys (agent37.com/dashboard/cloud/api-keys) or returned by POST /v1/signup; shown once. Each key is scoped to one workspace. Keys are not scoped: full access to the workspace. The signup docs say the key is “a full workspace key” that creates instances, manages templates and domains, funds the wallet and adds members; the only per-key restriction documented is an optional IP allowlist (up to 20 IPs/CIDRs). Sources: www.agent37.com: concepts, www.agent37.com: signup, www.agent37.com: members, www.agent37.com: exec, www.agent37.com: instance sizing.

Limits

  • 4 vCPU is the ceiling on every instance type; larger requests are clamped to 4 vCPU (memory up to 32 GB). (source)
  • Instances live until deleted; auto_sleep is off by default (idle_timeout_seconds default 900, 5 minutes to 1 day when enabled). (source)
  • Exec waits up to 15 minutes on sandbox.agent37.app and 13 minutes (780 s) on api.agent37.com, then errors while the command keeps running; CostGraph runs long commands in the background. Max 48 concurrent commands per instance. (source)
  • Instances run on dedicated servers in the United States only. (source)