Agent 37

Agent 37 offers a cloud sandbox platform for hosting AI agents. It supports persistent sandboxes and full Linux virtual machines to run code, services, and agents. The tool targets teams and developers who need hosted harnesses, tools, and credentials kept in isolated sandboxes, with ready-made harnesses or the option to bring your own. It emphasizes running multiple agent types in the cloud and managing environments centrally.

The platform operates by providing persistent sandboxes or VMs that can run Docker, Android emulators, and services reachable over HTTPS. An instance represents a persistent Linux computer, and a template determines what runs inside it. Users can start, stop, resize, or restart sandboxes, and connect to ports or signed previews. It supports bringing your own Docker image or using templates.

Key capabilities include hosting various agent harnesses (Hermes, OpenClaw, Claude Code, Codex, OpenCode, Grok, Pi), managed models, app connections (Gmail, Slack, Notion, GitHub, Google Calendar), and custom skills or plugins. The system tracks observability metrics, preserves files across sessions, and provides backups. It also supports auto-sleep to pause compute charges when idle and encrypted data at rest and in transit.

It highlights pricing with a 94% lower cost claim, a $5 starter credit for new accounts, and per-minute metering of compute and disk usage. It notes that standard sandboxes, VMs, and various templates are available, with references to sources, comparisons, and hosting options across browsers and API access.

Use cases

• Host AI agent harnesses in the cloud for organizations
• Run opensource agent templates like Hermes or OpenClaw