Agent 37

Product Introduction

1. Definition: Agent 37 is a managed cloud hosting platform providing persistent, always-on sandbox instances for autonomous AI agents such as Hermes, OpenClaw, and Claude Code. It is an API-driven infrastructure service specifically designed for developers and founders to deploy and manage isolated agent environments for their customers.
2. Core Value Proposition: It eliminates the need for developers to provision, manage, and scale individual VPS or Mac minis for each agent. Through a single API call, users can create dedicated, stateful agent instances where files, memory, and connected tools persist, enabling the rapid deployment of vertical AI products at a predictable, usage-based cost starting from $3.44/month.

Main Features

1. Persistent Agent Instances: Each customer receives their own dedicated, always-on computational instance. This sandbox maintains its own filesystem, memory, and tool connections until explicitly deleted, allowing for continuous, stateful agent operation across multiple sessions and API interactions.
2. Unified REST API & Instant Provisioning: The entire lifecycle—creation, messaging, command execution, and termination—is managed through a simple RESTful API. A single POST /v1/instances call with a specified template (e.g., agent37-hermes) and resource configuration (vCPU, RAM, Disk) instantly provisions a fully isolated environment, returning a unique ID and routable endpoint.
3. Kernel-Level Isolation with gVisor: Security and multi-tenancy are enforced at the kernel level using the gVisor container runtime. Each instance runs within its own hardened sandbox with a hard per-instance disk quota, ensuring strong isolation between different customers' agent environments on shared infrastructure.
4. Automatic Networking & Routing: Every created instance is automatically assigned a secure, HTTPS-enabled endpoint (e.g., https://{instance-id}.agent37.app). This eliminates the manual setup of reverse proxies, DNS records, and SSL certificates, providing immediate access to the agent's ports and streaming events.
5. Pay-for-Compute Billing Model: Costs are based on hourly usage of compute resources (vCPU, RAM, Disk) deducted from a prepaid balance. This model replaces per-seat licensing with a direct correlation to the actual infrastructure consumed per customer instance, with prorated refunds for unused time upon deletion.

Problems Solved

1. Pain Point: The operational overhead of manually provisioning, maintaining, and scaling stateful server environments (like VPSs or Mac minis) for every end-user's AI agent, which includes patching, monitoring, state tracking, and cost unpredictability.
2. Target Audience: SaaS Founders & Startup CTOs building AI-powered products who need to ship persistent agents to their clients; AI Application Developers creating tools for Hermes, OpenClaw, or custom Docker-based agents who require scalable, multi-tenant infrastructure without building glue code.
3. Use Cases: Shipping a white-label, customer-specific AI assistant platform; deploying per-client, always-on coding or research agents with their own project files; providing secure, isolated sandbox environments for agents that integrate with third-party services like Gmail, Slack, or Notion.

Unique Advantages

1. Differentiation: Unlike traditional IaaS (VPS) or FaaS (serverless functions), Agent 37 provides both persistent state and managed infrastructure. It abstracts away all DevOps tasks related to agent hosting, offering a purpose-built API for agent state management, whereas a VPS requires self-management and serverless lacks persistent filesystems.
2. Key Innovation: The core innovation is the "One API Call Per Customer" architecture. By combining persistent gVisor-based sandboxes with a unified REST API, metered compute billing, and automatic networking, it provides the simplest possible path for developers to convert a stateless application into a multi-tenant platform offering persistent, personalized AI agents.

Frequently Asked Questions (FAQ)

1. How does the pricing for Agent 37 compare to using AWS EC2 or DigitalOcean for multiple agent instances? Agent 37's pricing is designed for granular, per-instance cost control, starting at $3.44/month for a basic agent (2 vCPU, 4 GB RAM, 6 GB Disk). This all-inclusive price covers compute, API access, and networking, often representing a lower total cost than manually assembling equivalent EC2 instances with load balancers and storage, especially when accounting for DevOps time.
2. Can I bring my own Docker image and model API keys to Agent 37? Yes. The platform supports "Custom" templates where you can register any public Docker image. You can configure exposed ports and bring your own API keys for language models, or you can use the Agent 37 gateway to meter model usage directly from your prepaid balance with per-instance spend caps.
3. What security and isolation measures are in place for multi-tenant agent instances? Agent 37 uses gVisor, a user-space kernel, to provide strong, kernel-level isolation between each customer's agent instance. Every instance has its own dedicated disk quota and runs in an isolated sandbox, preventing cross-instance interference and ensuring secure multi-tenancy.
4. How do I interact with the agent after creating an instance? After provisioning via the API, you can message the agent using the POST /v1/instances/{id}/responses endpoint. You can also run shell commands, upload/download files, and subscribe to streaming events (like text or tool activity) over Server-Sent Events (SSE) for real-time updates, all authenticated via your API key.