Zeroboot

Stale
GitHub Rust Apache-2.0

Description

Sub-millisecond VM sandbox service based on Firecracker, providing hardware-isolated ultra-fast execution environments for AI agents via copy-on-write memory snapshots.

Key Features

  • Sub-millisecond startup — 0.79 ms p50 spawn latency via mmap MAP_PRIVATE, 1000 concurrent forks in 815 ms
  • Hardware-level isolation — Each fork is an independent KVM VM with hardware-enforced memory isolation
  • Minimal memory footprint — ~265 KB per sandbox, significantly better than E2B (~128 MB) and microsandbox (~50 MB)
  • Multi-language SDKs — Python and TypeScript SDKs with unified sandbox interface
  • Template prewarming — One-time boot preloads runtime, snapshots memory and CPU state for subsequent forks
  • REST API access — HTTP API endpoint accessible via API key without deployment

Use Cases

💡 Provide sub-millisecond startup isolated environments for AI agent code execution
💡 Support thousands of agents executing Python/JavaScript concurrently in high-throughput scenarios
💡 Provide safe sandboxing for LLM-generated code in production
💡 Quickly fork state during agent inference via CoW memory mechanism
💡 Deliver cloud code execution service orders of magnitude cheaper than traditional sandboxes

Tags

firecracker microvm sandbox sub-millisecond code-execution

Categories

🏗️ Sandbox & Execution

Quick Start

# Quick try via REST API
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))"}'
# Python SDK
pip install zeroboot
python3 -c "
from zeroboot import Sandbox
sb = Sandbox('zb_live_your_key')
result = sb.run('print(1 + 1)')
print(result)
"
Visit GitHub Visit Website View Docs

Related Projects