Overview
Agent 37 Cloud is managed infrastructure for products that need one persistent AI agent per customer. A single API call currently provisions a built-in Hermes template or a registered public custom Docker template with durable files, memory, tools, routable ports, and a REST interface. OpenClaw, Claude Code, and Codex system templates should be treated as coming or needs-verification unless the live template catalog shows them as available. The instance stays available until the developer deletes it, avoiding the usual work of operating a separate VPS and proxy for every user.
The service is aimed at founders building vertical-agent products rather than individual chatbot users. Each customer's agent runs under gVisor isolation with its own disk quota and optional spend cap. Developers can upload files, execute commands, stream reasoning and tool events, schedule work, and embed signed terminal, desktop, or file-browser sessions.
Agent 37 fits the AI agent category and can be compared with managed team-agent products such as WorkClaw. ToolWorthy's MCP beginner guide provides context for teams deciding how hosted agents should access external tools.
Key Features
- Per-customer instances - Provision a separate persistent computer, memory, files, and connected accounts for every end customer.
- Agent templates - Start with built-in Hermes templates such as
agent37-hermes-smalloragent37-hermes, or register a public custom Docker image and exposed ports; verify OpenClaw, Claude Code, and Codex template availability before listing them as active. - Unified REST API - Create instances, message agents, upload files, run commands, and stream events from a backend.
- Managed routing - Expose instance ports through authenticated URLs without configuring DNS, TLS, or a reverse proxy.
- Tool integrations - Connect Gmail, Slack, Notion, and other Composio-supported apps through in-agent OAuth authorization links; managed Composio calls are metered against the instance budget.
- Resource controls - Choose supported resource shapes and disk ranges, then set per-instance managed-spend budgets for LLM, Brave Search, and Composio calls.
How to Get Started
Create an API key, call /v1/instances with a template and resource allocation, then use the returned instance ID for messages and file operations. Start with the one-time $1 workspace credit, which is enough to run one small instance for several days, and set a low managed-spend cap before exposing the workflow to customers. Developers comparing agent communication standards may also read MCP versus A2A.
Pricing & Plans
Agent 37 bills prepaid compute by the hour. The official calculator lists $0.80 per vCPU per month, $0.70 per GB of RAM per month, and $0.09 per GB of disk per month.
| Example | Price |
|---|---|
| Starter credit | One-time $1 workspace credit, enough to run one small instance for several days |
| 1 vCPU, 3 GB RAM, 6 GB disk | About $3.44/month or $0.0047/hour for agent37-hermes-small after the starter credit is used |
| Additional disk | $0.09 per GB/month |
| Model usage | Bring your own keys, or use managed LLM calls billed at provider cost from the workspace wallet within a per-instance budget; the default managed-spend budget is $0 until raised |
Deleting an instance returns unused prepaid compute to the account balance on a prorated hourly basis.
Best For
- Founders shipping vertical AI agents to paying customers
- SaaS teams that need isolated agent state per tenant
- Developers hosting Hermes or OpenClaw without maintaining VPS fleets
- Products that need embedded terminals, desktops, or file access
FAQ
What does Agent 37 Cloud host?
It currently hosts built-in Hermes templates and registered custom Docker-based agents behind a managed API. OpenClaw, Claude Code, and Codex system-template availability should be verified against the live template catalog.
Is Agent 37 Cloud free?
New workspaces receive a one-time $1 credit, enough to run one small instance for several days. Ongoing instances use prepaid hourly compute billing from the workspace wallet.
How much does an always-on instance cost?
The 1 vCPU, 3 GB RAM, and 6 GB disk shape is approximately $3.44 per month for agent37-hermes-small; standard and custom templates start at 2 vCPU / 4 GB, about $4.94 per month.
Is customer data isolated?
Each instance runs under gVisor with its own persistent storage and disk quota.
Can developers use their own models?
Yes. Teams can bring model keys or use metered models through the Agent 37 balance.
Does it support custom agents?
Yes. Developers can register public custom Docker images and define the ports to expose.
