Repolode
Submit
zerobootdev

zerobootdev/zeroboot

Sub-millisecond KVM virtual machine sandboxes for running AI agent code, created by forking a pre-warmed Firecracker snapshot with copy-on-write memory.

Open-source alternative to:E2B

View on GitHubWebsite
Stars
2,468
Forks
107
Open issues
12
Last commit
20 days ago

What is it?

Zeroboot is a sandbox for running code generated by AI agents inside real virtual machines. Instead of booting a fresh VM for each execution, it takes a Firecracker snapshot of a VM that already has the runtime loaded, then forks new KVM VMs from that snapshot. Each fork maps the snapshot memory copy-on-write and restores the CPU state, which the project reports at about 0.8 ms per fork.

Each sandbox is a separate KVM virtual machine, so memory isolation is enforced by hardware rather than by process-level containment. The project provides a template step that boots a VM, preloads the runtime, and captures memory and CPU state, along with Python and TypeScript SDKs and an HTTP exec API. A managed API is also offered, and the code can be self-hosted on any Linux machine with KVM.

The project is described as a working prototype rather than production-hardened. Known limitations include a single vCPU per fork, no networking inside forks (communication is via serial I/O only), and shared CSPRNG state that requires userspace reseeding.

Key features

  • 01Forks new KVM VMs from a Firecracker snapshot in about 0.8 ms using copy-on-write memory.
  • 02Each sandbox is a separate VM with hardware-enforced memory isolation.
  • 03Provides Python and TypeScript SDKs plus an HTTP exec API.
  • 04Includes a one-time template step that preloads a runtime and snapshots its state.
  • 05Can be self-hosted on any Linux machine with KVM or used through the managed API.

Use cases

  • Running untrusted code generated by LLM agents with strong isolation.
  • Executing many short-lived code snippets where VM boot latency would dominate.
  • Spawning large numbers of parallel sandboxes for agent workloads.
  • Self-hosting a code execution backend on your own Linux infrastructure.

Quick start

  1. 1

    Try the managed API with a single curl request

    curl -X POST https://api.zeroboot.dev/v1/exec -H 'Content-Type: application/json' -H 'Authorization: Bearer zb_demo_hn2026' -d '{"code":"import numpy as np; print(np.random.rand(3))"}'
  2. 2

    Install the Python SDK from sdk/python

  3. 3

    Run a sample in Python with your API key

Things to know

  • Needs a Linux host with KVM to self-host.
  • Working prototype, not production-hardened yet.
  • Each fork is single vCPU and has no networking inside the sandbox.
  • Template updates require a full re-snapshot of about 15 seconds.

In collections

Sandboxes for AI agentsOpen-source AI agent frameworks & apps

Topics

#ai-agents#sandbox#virtual-machine#firecracker#kvm#copy-on-write#code-execution#rust#vm

FAQ

What is zeroboot?
Sub-millisecond KVM virtual machine sandboxes for running AI agent code, created by forking a pre-warmed Firecracker snapshot with copy-on-write memory. Zeroboot is a sandbox for running code generated by AI agents inside real virtual machines. Instead of booting a fresh VM for each execution, it takes a Firecracker snapshot of a VM that already has the runtime loaded, then forks new KVM VMs from that snapshot. Each fork maps the snapshot memory copy-on-write and restores the CPU state, which the project reports at about 0.8 ms per fork.
Is zeroboot free to use? What license does it have?
zeroboot is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is zeroboot still maintained?
Actively maintained. The last commit to zeroboot was on Oct 18, 2026.
What can I use zeroboot for?
Running untrusted code generated by LLM agents with strong isolation. Executing many short-lived code snippets where VM boot latency would dominate. Spawning large numbers of parallel sandboxes for agent workloads. Self-hosting a code execution backend on your own Linux infrastructure.
How do I install zeroboot?
Try the managed API with a single curl request: curl -X POST https://api.zeroboot.dev/v1/exec -H 'Content-Type: application/json' -H 'Authorization: Bearer zb_demo_hn2026' -d '{"code":"import numpy as np; print(np.random.rand(3))"}' Install the Python SDK from sdk/python Run a sample in Python with your API key
Where can I get zeroboot?
The source code and installation instructions are on GitHub: https://github.com/zerobootdev/zeroboot

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects