agenthelm-sdk (v0.5.0)

Monitor your AI agents from AgentHelm. Add one line. See everything.

Install

npm install agenthelm-node-sdk

Features

• 🛡️ Safety Firewall: Classify actions as read, sideEffect, or irreversible (Human-in-the-Loop).
• ⚡️ One-line Integration: Import AgentHelm and you're live.
• 📊 Token Tracking: Real-time monitoring of LLM costs (INR/USD).
• 🛰️ Integrity Checkpoints: Save/Resume agent state with built-in resumability.
• 🤝 Handshake Protocol: Secure, JWT-based authentication.

Usage (Standard)

import { connect } from 'agenthelm-node-sdk'

const dock = connect({
  key: 'ahe_live_xxxxx',
  name: 'My Agent'
})

dock.log('Agent started')
dock.trackTokens({ used: 1500, model: 'gemini-flash' })

🛡️ Safety Firewall (Classification-First)

Ensure mission-critical tasks are approved by a human before they execute.

// 1. Read-only (Always safe, no gating)
const data = await agent.read(() => fetch_db_record(id));

// 2. Side Effect (Logs and retries, but non-blocking)
await agent.sideEffect(() => send_notification(user), { maxRetries: 5 });

// 3. Irreversible (BLOCKS until approved via Telegram or Dashboard)
await agent.irreversible(async () => {
  return await stripe.charges.create({ ... });
}, { timeout: 60000 });

API

connect(options) → AgentHelm instance

Options:

| Option | Type | Description | |---|---|---| | key | string | Your connect key (required) | | name | string | Agent display name | | agentType | string | python | node | other | | version | string | Your agent version | | burnRateThreshold| number | Alert if tokens/min exceeds this |

Safety Methods:

• dock.read(task)
• dock.sideEffect(task, options?)
• dock.irreversible(task, options?)

Standard Methods:

• dock.log(message, level?, data?)
• dock.output(data, label?)
• dock.error(message, error?)
• dock.trackTokens({ used, model, costPer1k? })
• dock.checkpoint(stepName, state, options?)
• dock.resumeFrom(taskId)
• dock.stop()
• dock.listen()

Documentation

Full docs at https://agenthelm.online