Authensor quickstart guide

This quickstart gets you enforcing safety policies on agent tool calls as fast as possible. No control plane, no database, no infrastructure. Just the SDK and a policy file.

1. Install

pnpm add @authensor/sdk

2. Create a policy

Save this as policy.yaml:

version: "1"
rules:
  - tool: "file.read"
    action: allow

  - tool: "file.write"
    action: escalate
    reason: "File writes need approval"

  - tool: "shell.*"
    action: block
    reason: "Shell access is disabled"

3. Guard your tool calls

import { createGuard } from '@authensor/sdk';

const guard = createGuard({ policyPath: './policy.yaml' });

// This goes through
const read = guard('file.read', { path: '/tmp/data.json' });
console.log(read.action); // 'allow'

// This gets escalated
const write = guard('file.write', { path: '/etc/config', content: '...' });
console.log(write.action);  // 'escalate'
console.log(write.reason);  // 'File writes need approval'

// This gets blocked
const shell = guard('shell.execute', { command: 'whoami' });
console.log(shell.action); // 'block'
console.log(shell.reason); // 'Shell access is disabled'

4. Add content scanning

Install Aegis and enable it in your guard:

pnpm add @authensor/aegis

const guard = createGuard({
  policyPath: './policy.yaml',
  aegis: { enabled: true }
});

Now every tool call is also scanned for prompt injection patterns, exposed credentials, and PII before the policy engine evaluates it. If Aegis detects a threat, the action is blocked regardless of what the policy says.

5. Check your receipts

Every guard call generates a receipt. Receipts form a hash chain where each receipt includes the hash of the previous one:

const decision = guard('file.read', { path: '/tmp/data.json' });
console.log(decision.receipt);
// {
//   id: "rec_abc123",
//   tool: "file.read",
//   action: "allow",
//   timestamp: "2026-01-15T10:30:00Z",
//   hash: "sha256:...",
//   previousHash: "sha256:..."
// }

What comes next

• Approval workflows: Handle escalate decisions with human-in-the-loop review
• Sentinel monitoring: Track behavioral patterns over time
• Control plane: Centralized policy management with a REST API
• Framework adapters: Drop-in integrations for LangChain, OpenAI Agents SDK, and CrewAI