@vaiuai/telemetry

Vaiu Telemetry SDK for Node.js and TypeScript applications. Instruments LLM calls, tracking model traces, latency, and token usage to send to Vaiu Telemetry.

Installation

npm install @vaiuai/telemetry

Quick Start

Environment Variables

Set environment variables:

VAIU_API_HOST=https://your-telemetry-host.com
VAIU_API_PUBLIC_KEY=pk-vaiu-...
VAIU_API_SECRET_KEY=sk-vaiu-...

Usage

import { VaiuTelemetry } from "@vaiuai/telemetry";

const client = new VaiuTelemetry(); // loads from env vars

// 1. Create a Trace
const trace = client.trace({
  name: "chat-workflow",
  userId: "user-123",
});

// 2. Start a Generation
const generation = trace.generation({
  name: "openai-gpt-4o",
  model: "gpt-4o",
  input: "Hello, world!",
});

// ... call your model API ...

// 3. End Generation
generation.end({
  output: "Hello! How can I help you today?",
  usage: {
    input: 10,
    output: 15,
    total: 25,
  },
});

// 4. Flush Telemetry
await client.flushAsync();

Convenient Helpers

// Observe async function
const result = await client.observe(
  "doc-summary",
  "gpt-4o-mini",
  async () => {
    return await myLlmCall();
  },
  (res) => ({
    output: res.text,
    usage: { input: res.promptTokens, output: res.completionTokens },
  })
);

// One-shot generation logging
await client.logGeneration({
  name: "completion",
  model: "claude-3-5-sonnet",
  latencyMs: 450,
  promptTokens: 100,
  completionTokens: 50,
  input: "Prompt string",
  output: "Completion string",
});