@agentskit/sandbox

Let agents write and run code safely — in isolated cloud VMs, not on your machine.

Tags: ai · agents · llm · agentskit · ai-agents · sandbox · code-execution · e2b · secure-execution · code-interpreter

Why sandbox

• Code generation that actually executes — agents can write, run, and iterate on code without you worrying about what they'll do to your filesystem or OS
• E2B cloud VMs out of the box — each execution runs in an isolated environment with configurable timeouts, no network by default, and a 50MB memory cap
• Bring your own backend — the SandboxBackend interface is 2 methods; plug in Docker, Firecracker, or any custom isolation layer
• Works alongside any other tools — add sandboxTool to the same tools array as webSearch or filesystem; no special wiring needed

Install

npm install @agentskit/sandbox @e2b/code-interpreter

Quick example

import { createRuntime } from '@agentskit/runtime'
import { anthropic } from '@agentskit/adapters'
import { sandboxTool } from '@agentskit/sandbox'

const runtime = createRuntime({
  adapter: anthropic({ apiKey: process.env.ANTHROPIC_API_KEY, model: 'claude-sonnet-4-6' }),
  tools: [sandboxTool({ apiKey: process.env.E2B_API_KEY })],
})

const result = await runtime.run('Write and run a Python script that generates a Fibonacci sequence up to 100')
console.log(result.content)

Features

• sandboxTool({ apiKey }) — drop-in tool for code execution via E2B cloud VMs
• Configurable timeouts and resource limits
• No network by default — agents cannot exfiltrate data
• SandboxBackend interface — 2 methods to bring Docker, Firecracker, or any custom backend
• Follows ToolDefinition contract — works in runtime, useChat, or any custom loop

Ecosystem

| Package | Role | |---------|------| | @agentskit/runtime | createRuntime({ tools }) | | @agentskit/tools | code tool can delegate to sandbox | | @agentskit/adapters | LLM for codegen tasks | | @agentskit/core | Tool contract |

Contributors

License

MIT — see LICENSE.

Docs

Full documentation · GitHub