Agent 37
Agent 37 provides secure, cost-effective cloud sandbox infrastructure to host, run, and scale AI agent harnesses and background workloads.
What is Agent 37?
Agent 37 is a dedicated cloud sandbox platform designed to run AI agents, harnesses, and background workloads. Built to eliminate the operational burden of managing server fleets, the service allows developers to deploy existing agent harnesses such as Claude Code, Codex, OpenClaw, and Hermes, or bring custom Docker images. Workloads run inside kernel-enforced sandboxes with dedicated filesystems, persistent storage, and outbound-only network security.
How does Agent 37 work?
- Choose a ready-made agent harness or supply a custom Docker image to define your environment.
- Configure compute specifications, storage volumes, and optional integrations or managed models.
- Deploy sandboxes with authenticated URLs, public ports, or custom domain mappings.
- Run workloads with automatic sleep triggers that pause compute billing during idle periods while retaining disk state.
- Monitor execution through built-in observability tools tracking CPU, memory, disk usage, and application logs.
Use Cases
- Hosting customer-facing AI coding assistants and live development environments in isolated sandboxes.
- Running persistent grant fundraising, research, or workflow automation agents unattended on cron schedules.
- Powering multi-tenant agent applications where each tenant requires dedicated filesystem storage and separate credentials.
Key Features
- Kernel-level sandbox isolation powered by gVisor with dedicated filesystems and encrypted storage.
- Metered pay-as-you-go pricing billed per minute of compute with auto-sleep to pause idle costs.
- Persistent disk storage and automatic backup support across restarts and sleep cycles.
- Support for popular agent harnesses, custom Docker images, and managed LLM model keys.
- Built-in observability tracking CPU, memory, disk usage, and execution logs.
Frequently Asked Questions
How does billing work on Agent 37?
Agent 37 operates on a prepaid usage model without subscription fees. Sandboxes are metered per minute for compute uptime, while disk storage is charged separately, and new accounts receive a $5 starter credit once a payment card is added.
What happens to sandbox data when an agent is idle or asleep?
When an agent is idle or enters auto-sleep mode, compute billing stops immediately and only disk storage is charged, keeping all workspace files and memory preserved for the next session.
Can I use my own model API keys and Docker containers?
Yes, Agent 37 supports custom Docker images and allows users to either use managed model endpoints or bring their own API keys without deducting from their platform balance.
How are workloads isolated and secured?
Every sandbox is isolated using kernel-level gVisor enforcement with an independent filesystem, encrypted storage at rest, Cloudflare TLS ingress requiring API keys, and outbound-only network egress.
Alternative Tools
-
Bagasura A bug tracking platform that streamlines workflows and enhances collaboration for increased productivity. -
BashSenpai Quickly convert instructions into commands for your terminal assistant. -
Chariot A platform that helps developers add advanced NLP features into their applications.
-
Floneum Create efficient AI workflows with WebAssembly plugins. -
Pinecone "Secure, scalable vector search database—powerful solution for high-performance applications."
-
Proddy.io A powerful documentation generator for creating product documentation.