Agent 37 Cloud

Agent 37 Cloud is a managed infrastructure platform that lets developers provision persistent, isolated agent sandboxes for each of their customers with a single API call. Built by the Agent 37 team, it supports Hermes, OpenClaw, and any custom Docker image, wrapping each agent's CLI or SDK in a clean REST API with streaming. The platform claimed the #1 Product of the Day on Product Hunt, according to the Agent 37 website.

What It Is

Agent 37 Cloud sits in the agent-hosting layer of the AI stack. Instead of renting VPSs, managing containers, and wiring reverse proxies, developers POST to a single endpoint and get back a running instance with a routable URL, persistent files, memory, and connected tools. Each instance maps to one customer and stays alive until explicitly deleted. The platform targets product teams that want to ship multi-tenant agent products without building or operating the underlying infrastructure.

Architecture: One Instance Per Customer

The core design principle is strict per-customer isolation. A single POST /v1/instances call provisions a sandbox with configurable vCPU, RAM, and disk. Every port of every instance is immediately routable at https://{id}.agent37.app, authenticated behind the caller's API key or a signed URL — no reverse proxy, DNS records, or TLS certificates to manage. Instances run under the gVisor runtime with kernel-level isolation and a hard per-instance disk quota.

Templates and Customization

Agent 37 Cloud ships with a catalog of pre-built templates:

• Hermes — the default template with chat, browsing, code, and file capabilities
• OpenClaw — a channels-and-models agent hosted behind the same API
• Claude Code — listed as "coming soon" on the site
• Custom — any public Docker image, registered under a name, with configurable exposed ports

Developers can preinstall their own skills, prompts, and tools via a custom template, enabling white-label deployments where end customers never see the Agent 37 brand.

Capabilities Included in Every Instance

The platform bundles a broad set of agent-product primitives so teams don't need to deploy additional services:

• Any model via a built-in gateway, or bring-your-own API keys with per-instance spend caps
• Integrations with Gmail, Slack, Notion, and 1,000+ apps via Composio, routed and metered per instance
• Files — upload to and download from instances directly from a backend
• Shell commands — run arbitrary commands inside an instance from the backend
• Streaming events — text, reasoning, and tool activity over Server-Sent Events
• Scheduled runs — cron-based unattended execution with readable results
• Embedded UI — live terminal, desktop (via noVNC), and file browser accessible via signed URLs

Deployment Model and Billing

Agent 37 Cloud uses a prepaid, usage-based billing model. Compute is priced per vCPU, per GB of RAM, and per GB of disk per month, billed hourly. Unused prepaid time returns to the balance prorated to the hour when an instance is deleted. Developers can set per-instance spend caps and bring their own model API keys or meter models from the same balance. The managed hosting tier (agent37.com root) offers fixed monthly plans with bundled resources, while the Cloud API tier (agent37.com/cloud) is purely usage-based compute.

White-Label and Reseller Path

Agent 37 Cloud is explicitly designed for white-label deployments. The platform page states that customers see only the builder's product, never Agent 37. Resellers can provision agents through the API, preload skills and prompts, and package ready-to-run agents for specific verticals, then price their own support, onboarding, and workflows on top of the underlying compute cost.