LLM Lens is a blazing-fast, developer-first alternative to cloud AI observability platforms. It intercepts traces at runtime and pipes them directly to a local, WAL-optimized SQLite daemon with < 1ms latency.

This package (llm-lens-sdk) is the lightweight OpenTelemetry wrapper that instruments your LLM frameworks.

Installation

npm install llm-lens-sdk

(Note: You must also run the llm-lens-daemon to receive the traces and view the dashboard).

Usage

LLM Lens integrates seamlessly with popular AI frameworks. First, initialize the telemetry exporter:

Vercel AI SDK (Native OpenTelemetry)

Vercel AI SDK works out-of-the-box using its built-in experimental_telemetry option. initAutopilot automatically registers a fast, lightweight OpenTelemetry provider.

import { initAutopilot } from 'llm-lens-sdk';
import { generateText } from 'ai';

// 1. Initialize telemetry (registers global OTEL provider automatically)
initAutopilot({ serviceName: 'my-vercel-ai-agent' });

// 2. Just use Vercel AI SDK natively with telemetry enabled!
await generateText({
  model,
  prompt: 'Hello world',
  experimental_telemetry: { isEnabled: true } // Vercel handles the rest!
});

OpenAI (and Groq)

import { initAutopilot, wrapOpenAI, wrapGroq } from 'llm-lens-sdk';
import OpenAI from 'openai';
import Groq from 'groq-sdk';

initAutopilot({ serviceName: 'my-openai-agent' });

// Wrap the client instance
const openai = wrapOpenAI(new OpenAI({ apiKey: '...' }));
const groq = wrapGroq(new Groq({ apiKey: '...' }));

// ... use openai.chat.completions.create as normal ...

For more documentation, visit the main repository.