deeplethe

deeplethe / forkd

Public

Fork microVMs sandbox from a warmed parent in 101 ms.

ai-agents copy-on-write kvm microvm rust
20
1
100% credibility
Found May 13, 2026 at 29 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Rust
AI Summary

forkd is a microVM runtime that enables forking hundreds of isolated Linux environments from a single warmed parent snapshot for fast AI agent sandboxes.

How It Works

1
🔍 Discover forkd

You hear about a clever way to create hundreds of safe, isolated spaces for AI helpers to run code super quickly, without slow startups.

2
🛠️ Prepare your setup

You run a simple preparation script to get your computer ready for creating these fast spaces.

3
📦 Build a ready workspace

You pick or create a prepared environment with all the tools your AI needs, like Python libraries, already loaded.

4
💾 Capture the warm starting point

You boot one space, let it warm up with everything ready, and save this perfect snapshot to reuse instantly.

5
🚀 Launch many copies at once

With one command, you instantly fork dozens or hundreds of identical spaces, each fully isolated but sharing the warm setup.

6
🤖 Run AI tasks safely

Your AI helpers now run code, tests, or analyses in these secure spaces, feeling lightning-fast thanks to the shared warm state.

🎉 Scale effortlessly

You achieve blazing speeds for fan-out tasks like code interpreters, with strong isolation and no cold starts—your AI workflows fly!

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 29 to 20 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is forkd?

forkd forks microVM sandboxes from a pre-warmed parent VM snapshot in 101 ms, delivering KVM isolation with copy-on-write memory sharing. Built in Rust on Firecracker, it boots a parent once—loading Python deps, ML models, or JIT caches—then spawns isolated children that inherit the warm state for free. Developers get a CLI for snapshots and forks, a REST API daemon with auth/audit/metrics, and an E2B-compatible Python SDK.

Why is it gaining traction?

It crushes cold-start alternatives: 101 ms for 100 sandboxes vs Docker's 335 s or Firecracker's 759 ms, with just 0.12 MiB host memory delta per child. Recipes for numpy, Jupyter, Node.js, or E2B code interpreters skip per-request imports, perfect for fan-out. Like forking a GitHub repo to GitLab but for VMs—real Linux kernels, networking, and cgroups per child, no vendor lock-in.

Who should use this?

AI agent builders running code interpreters or tool-calling rollouts, like Anthropic-style evals or SWE-bench harnesses needing git/pytest in parallel sandboxes. Eval teams fanning out Jupyter kernels or Node.js tests without Docker overhead. Self-hosters replacing E2B SaaS for untrusted code execution in CI.

Verdict

Grab it for agent fan-out prototypes—benchmarks deliver, recipes accelerate starts. Alpha with 20 stars and 1.0% credibility means APIs may shift pre-1.0; test locally first.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.