@declaw/sdk

Secure runtime for AI agents. Spin up isolated sandboxes in milliseconds with built-in guardrails — PII scanning, prompt injection defense, network isolation, and egress filtering.

Install

npm install @declaw/sdk

Quick Start

import { Sandbox } from '@declaw/sdk';

const sandbox = await Sandbox.create({
  apiKey: 'your-api-key',
  template: 'base',
  timeout: 60,
});

// Run commands
const result = await sandbox.commands.run('echo "Hello from a secure sandbox"');
console.log(result.stdout);

// Read/write files
await sandbox.files.write('/tmp/hello.txt', 'Hello World');
const content = await sandbox.files.read('/tmp/hello.txt');

// Clean up
await sandbox.kill();

Why Declaw?

AI agents need to execute code, call APIs, and interact with the world. Declaw gives them a secure sandbox to do it — with built-in guardrails that protect your users and infrastructure.

• Sub-10ms sandbox creation — pre-warmed VM pool, no cold starts
• Network isolation — per-sandbox firewall with domain and CIDR rules
• Full file system — read, write, upload, download files in the sandbox

Security & Guardrails

Every outbound request from the sandbox passes through a configurable security pipeline.

PII Scanning

Detect and redact sensitive data before it leaves the sandbox.

const sandbox = await Sandbox.create({
  security: SecurityPolicy.from({
    pii: {
      enabled: true,
      types: ['ssn', 'credit_card', 'email', 'phone', 'api_key'],
      action: 'redact',
    },
  }),
});

Prompt Injection Defense

Block prompt injection attempts in agent outputs.

const sandbox = await Sandbox.create({
  security: SecurityPolicy.from({
    injectionDefense: {
      enabled: true,
      action: 'block',
      threshold: 0.85,
    },
  }),
});

Toxicity Filtering

security: SecurityPolicy.from({
  toxicity: { enabled: true, action: 'block', threshold: 0.7 },
})

Code Security & Invisible Text Detection

security: SecurityPolicy.from({
  codeSecurity: { enabled: true, action: 'log' },
  invisibleText: { enabled: true, action: 'block' },
})

Network Policies

// Allow only specific domains
const sandbox = await Sandbox.create({
  network: { allowOut: ['api.openai.com', 'huggingface.co'] },
});

// Block all egress
const isolated = await Sandbox.create({
  network: { denyOut: ['ALL_TRAFFIC'] },
});

Data Transformation

Transform sensitive values in-flight.

security: SecurityPolicy.from({
  transformations: [
    { pattern: 'sk-[a-zA-Z0-9]+', replacement: '[API_KEY]', direction: 'egress' },
  ],
})

Combining Guardrails

All guardrails compose — enable multiple and they run in sequence:

const sandbox = await Sandbox.create({
  template: 'ai-agent',
  timeout: 300,
  network: { allowOut: ['api.openai.com', 'api.anthropic.com'] },
  security: SecurityPolicy.from({
    pii: { enabled: true, action: 'redact', types: ['ssn', 'credit_card'] },
    injectionDefense: { enabled: true, action: 'block' },
    toxicity: { enabled: true, action: 'log' },
    invisibleText: { enabled: true, action: 'block' },
  }),
});

Templates

| Template | Description | |----------|-------------| | base | Minimal Linux | | python | Python 3.12 with pip | | node | Node.js 22 LTS with npm | | code-interpreter | Python with data science libraries | | ai-agent | Python + Node.js + AI/ML tools | | mcp-server | MCP server runtime | | web-dev | Node.js + browser testing | | devops | Docker, Terraform, kubectl |

API

// Create sandbox
const sandbox = await Sandbox.create({ template, apiKey, timeout, network, security });

// Commands
const result = await sandbox.commands.run('ls -la');
const stream = sandbox.commands.stream('python script.py');

// Files
await sandbox.files.write(path, content);
const data = await sandbox.files.read(path);
const entries = await sandbox.files.list('/');

// PTY (interactive terminal)
const pty = await sandbox.pty.create({ cols: 80, rows: 24 });

// Lifecycle
await sandbox.kill();

License

Apache-2.0