agentlock-sdk-node

Node.js SDK for integrating AI agents with AgentLock.

Installation

pnpm add agentlock-sdk-node
# or: npm i agentlock-sdk-node / yarn add agentlock-sdk-node

Quick start

import { AgentLockClient, generateKeypair } from 'agentlock-sdk-node';

const client = new AgentLockClient({
  baseUrl: 'https://app.agentlock.net',
  agentId: process.env.AGENT_ID!,
  privateKey: process.env.AGENT_PRIVATE_KEY!,
});

const result = await client.requestAction({
  action_type: 'write',
  tool: 'http',
  payload: {
    url: 'https://httpbin.org/post',
    method: 'POST',
    body: { message: 'Hello from Node!' },
  },
});

if (result.status === 'PENDING') {
  const final = await client.awaitResult(result.request_id);
  console.log(final);
}

generateKeypair() produces an Ed25519 keypair — register the public key in the dashboard and keep the private key in your agent environment.

SSH sessions

For multi-command workflows, open a session once and reuse it:

const session = await client.ssh.open({ credentialId: 'cred-1' });
try {
  await session.run('uname -a');
  await session.run('uptime');
} finally {
  await session.close();
}

session.close() is idempotent — safe to call from finally.

To re-attach to a session opened in another process:

const sessions = await client.ssh.list();
const s = sessions.find(s => s.host === 'ops-prod-01');
if (s) await s.run('tail -n 50 /var/log/nginx/access.log');

client.sshRun() (one-shot) is @deprecated but still supported for true single-command use — it sends ssh.run directly so policy commandRules can still auto-allow. For multi-command workflows, use client.ssh.open() to amortize the connection cost.

Browser snapshot

const snap = await client.browserSnapshot(sessionId);
// { html, url, title, screenshot? }

HTTP requests

const res = await client.httpRequest({
  method: 'POST',
  url: 'https://api.example.com/items',
  body: { name: 'x' },
  credentialId: 'cred-api',
});
// { status, headers, body, durationMs }