runtime-monitor-sdk

JavaScript / TypeScript SDK for Runtime Monitor — real-time cost guardrails, kill-switches, and hallucination detection for AI agents.

Install

npm install runtime-monitor-sdk

Quick start

import { AgentMonitor } from "runtime-monitor-sdk";

const monitor = new AgentMonitor({
  apiKey: "am_...",
  agentName: "my-agent",
  budgetLimitUsd: 1.00,
});

await monitor.start();

// Track individual LLM calls manually
monitor.track({
  model: "gpt-4o",
  inputTokens: 1200,
  outputTokens: 400,
});

await monitor.stop();

Auto-wrap OpenAI client

import OpenAI from "openai";
import { AgentMonitor } from "runtime-monitor-sdk";

const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

const monitor = new AgentMonitor({
  apiKey: "am_...",
  agentName: "my-agent",
  budgetLimitUsd: 2.00,
});

await monitor.start();

// All calls through wrappedOpenAI are tracked automatically
const wrappedOpenAI = monitor.wrapOpenAI(openai);
const response = await wrappedOpenAI.chat.completions.create({
  model: "gpt-4o",
  messages: [{ role: "user", content: "Hello" }],
});

await monitor.stop();

Kill-switch

monitor.onKillSwitch((reason) => {
  console.error("Budget exceeded:", reason);
  process.exit(1);
});

Run a function with automatic session lifecycle

const result = await monitor.run(async () => {
  // your agent logic here
  return "done";
});

API

new AgentMonitor(options)

| Option | Type | Required | Description | |---|---|---|---| | apiKey | string | ✅ | Your Runtime Monitor API key (am_...) | | agentName | string | ✅ | Name displayed in the dashboard | | budgetLimitUsd | number | — | Hard budget cap; triggers kill-switch when exceeded | | baseUrl | string | — | Override API URL (default: https://monitor.global-soft.ro) | | metadata | object | — | Arbitrary key/value pairs attached to the session |

Methods

| Method | Description | |---|---| | start() | Create a new session on the server | | stop() | End the session and flush pending data | | track(options) | Record a single LLM call | | run(fn) | Start session → run fn → stop session, returns fn result | | wrapOpenAI(client) | Returns a proxy that auto-tracks every completion call | | onKillSwitch(cb) | Register callback invoked when the budget is exceeded |

Python SDK

pip install agent-monitor

See the main repo README for full documentation.

License

MIT