Skip to main content

​
What is Hatch?

Hatch is a wrapper around Firecracker for spinning up microVMs, designed for agentic workloads. It provides a REST API for lifecycle management, wake-on-request, snapshot/restore for idle VMs, and subdomain-based reverse proxy.

Quickstart

Get up and running with Hatch in minutes

Installation

Detailed setup and configuration guide

​
Key Features

REST API

Full lifecycle management via HTTP API - create, stop, snapshot, restore, and delete VMs with simple curl commands

Wake-on-Request

Freeze idle VMs to zero compute and wake them transparently on HTTP requests or SSH connections

Snapshot/Restore

Save VM state to S3-compatible storage and restore VMs to exactly where they were paused

Subdomain Routing

Route HTTP traffic to VMs by subdomain with automatic wake for snapshotted instances

SSH Forwarding

Dedicated host ports for SSH access with automatic wake-on-SSH for snapshotted VMs

Idle Auto-Snapshot

Automatically snapshot idle VMs to save compute resources

​
Architecture Overview

Hatch orchestrates Firecracker microVMs with a full networking stack, reverse proxy, and snapshot management. Hatch Architecture

​
Core Components

VM Manager
Manages the Firecracker process lifecycle, networking setup (bridge, TAP devices), and resource allocation. API Server (:8080)
HTTP API for VM lifecycle operations - create, list, stop, delete, snapshot, restore, and route management. Reverse Proxy (:9090)
Subdomain-based routing to VMs with wake-on-request support. Requests to my-agent.hatch.local are proxied to the corresponding VM. SSH Gateway
Port forwarding for SSH access with wake-on-SSH capability. Each networked VM gets a dedicated host port. Idle Monitor
Background loop that tracks VM activity and automatically snapshots idle VMs to S3. Snapshot Storage
S3-compatible storage for VM snapshots (memory, CPU state, and disk delta).

​
Use Cases

​
Agentic Workloads

Hatch is designed for AI agents that need isolated, ephemeral compute environments:
  • Development agents that need a full Linux environment to run code and tests
  • Task automation requiring isolated execution contexts
  • Multi-tenant workloads where each user gets their own VM
  • On-demand compute that scales to zero when idle

​
Serverless VMs

The wake-on-request pattern enables serverless-style VMs:
  • Idle VMs are snapshotted and frozen (zero compute cost)
  • HTTP requests or SSH connections automatically wake VMs
  • VMs resume from exactly where they were paused
  • No cold start - VMs restore with full state intact

​
How It Works

1

Create a VM

POST to /vms with configuration (CPU, memory, network, cloud-init user data)
2

Networking is set up

Hatch creates a TAP device, allocates IP/MAC, configures DHCP, and sets up NAT
3

Firecracker starts the VM

VM boots with cloud-init configuration and connects to the network
4

Access via SSH or HTTP

Connect via dedicated SSH port or register HTTP routes for subdomain access
5

Idle detection

After inactivity timeout, VM is automatically snapshotted to S3 and stopped
6

Wake on next request

Next SSH or HTTP request restores the VM from S3 and forwards the connection

​
Next Steps

Quickstart

Create your first VM in 5 minutes

Installation Guide

Full installation and configuration details

Build docs developers (and LLMs) love

Get started for freeTalk to us