Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/deeplethe/forkd/llms.txt

Use this file to discover all available pages before exploring further.

forkd is a microVM sandbox runtime built for AI agent fan-out. Instead of cold-booting each sandbox, forkd boots a parent VM once, warms it up (imports, JIT compilation, model weights), and then forks children via copy-on-write — each child inherits the parent’s memory instantly, just like fork(2) but with full KVM isolation per child.

Quickstart

Zero to forking microVMs in one command — with forkd installed and running in minutes.

How It Works

Understand the copy-on-write snapshot model, BRANCH mechanics, and the controller daemon.

REST API

Full HTTP API reference for snapshots, sandboxes, branching, and metrics.

Python SDK

E2B-compatible Python SDK — spawn, branch, exec, and eval against live microVMs.

​
Why forkd?

When an AI agent fan-out spawns 100 sandboxes, the bottleneck isn’t compute — it’s the time each sandbox spends re-importing numpy, re-loading a model, or re-warming a JIT. forkd eliminates that cost entirely: the parent VM does it once, and every child inherits the result for free.

100 VMs in 101 ms

Fork 100 KVM-isolated microVMs from one warm snapshot in ~101 ms wall-clock — 10× faster than the next-fastest open-source alternative.

BRANCH in 56 ms

Pause a running agent mid-thought, snapshot its state, and resume in sub-50 ms using UFFD_WP live mode (v0.4).

Real Linux per child

Each child is a full Firecracker microVM: multi-vCPU, real networking, apt install, outbound HTTPS — not a restricted function sandbox.

Diff-snapshot chains

Stack pip-install layers as linked diff snapshots (v0.5) — share the base, pay only for deltas at spawn time.

​
Get started

1

Install forkd

Download the pre-built binary tarball — no Rust toolchain required.
curl -sSL https://github.com/deeplethe/forkd/releases/download/v0.5.2/forkd-v0.5.2-x86_64-linux.tar.gz \
  | sudo tar -xz -C /usr/local/bin/
2

Run quickstart

Preflight your host, bake a Python snapshot, and fork your first 10 microVMs.
sudo -E forkd quickstart
3

Fork at scale

Once a snapshot is ready, fork 100 children in one command.
sudo -E forkd fork --tag quickstart -n 100 --per-child-netns
4

Drive from Python or TypeScript

Use the SDK to spawn, branch, and execute code inside sandboxes from your agent framework.
from forkd import Controller
c = Controller()
children = c.spawn_sandboxes("pyagent", n=10, per_child_netns=True)

​
Explore the docs

Host Setup

Configure KVM, tap devices, and network namespaces on your Linux host.

Building Snapshots

Boot a parent VM, warm it up, and create a reusable snapshot tag.

Branching

BRANCH a live running sandbox mid-thought in sub-50 ms.

Snapshot Chains

Stack diff snapshots to share a base across multiple pip-installed layers.

Recipes

Drop-in patterns for LangGraph, CrewAI, AutoGen, OpenAI Swarm, and more.

Operations

Run the controller daemon, set up auth, scrape metrics, and manage audit logs.

Build docs developers (and LLMs) love

Get started for freeTalk to us