Agent37 Cloud hosts agents as persistent, always-on sandboxes behind a single API. You create an instance from a template, such as Hermes, OpenClaw, Claude Code, or a custom harness, and then message it; the platform is aimed at products that want to hand each of their own users a dedicated agent without running the infrastructure themselves.

A POST /v1/instances call with a template name returns an instance ID, its status, and exposed ports, each with a public HTTPS URL under agent37.app. Files, memory, and connected tools stay on the instance until you delete it, so an agent keeps its state across conversations. Instances support chat, browsing, code execution, file operations, shell access, streaming events, custom skills, and scheduled runs.

Billing is prepaid and metered: compute is charged by the minute only while a sandbox is awake, and disk is charged from creation until deletion. Sandboxes are isolated with gVisor, routed through Cloudflare's edge, egress-firewalled, and encrypted at rest and in transit. A white-label option lets you resell OpenClaw and Hermes under your own brand.

Resource	Rate
vCPU	$0.0011 per vCPU-hour
Memory	$0.00096 per GB-hour
Disk	$0.00012 per GB-hour

Shared size	vCPU	Memory
Small	2	4 GB
Medium	4	8 GB
Large	8	16 GB

Features

Template instances: Hermes, OpenClaw, Claude Code, or your own harness as a running sandbox
One API: create an instance, then send messages and receive streaming events
Persistent state: files, memory, and tool connections survive across sessions until deletion
Per-instance URLs: each exposed port gets a public HTTPS address
Agent capabilities: browsing, code, files, shell, custom skills, and scheduled runs
Metered billing: compute while awake, disk from creation to deletion, no plan fee
Isolation: gVisor kernel sandboxing, egress firewall, and encryption at rest and in transit
White-label: resell hosted OpenClaw and Hermes instances under your own brand
Dedicated sandboxes: available beyond the three shared sizes

Claim Agent37 Cloud