lumni-ai

AI agent observability SDK — track interactions, auto-instrument LLM calls, collect feedback signals.

Install

npm install lumni-ai

Quick Start

import * as lumni from "lumni-ai";

lumni.init("your-ingest-key", { baseUrl: "https://your-backend.example.com" });

// Auto-instrument all LLM calls (OpenAI, Anthropic, Gemini)
// Works with LangChain.js, Vercel AI SDK, or any framework using these providers.

Track an AI Interaction

const eventId = lumni.trackAi({
  userId: "user-123",
  event: "chat_response",
  model: "gpt-4o",
  input: "What's the weather?",
  output: "It's sunny and 72F.",
});

Auto-Instrumentation (Zero Code)

import * as lumni from "lumni-ai";
lumni.init("key", { baseUrl: "http://localhost:8085", tracingEnabled: true });

// Every OpenAI/Anthropic/Gemini call is now captured automatically
import OpenAI from "openai";
const client = new OpenAI();
await client.chat.completions.create({ model: "gpt-4o", messages: [...] }); // tracked

User Identification

lumni.identify("user-123", { name: "Alice", email: "[email protected]", plan: "paid" });

Feedback Signals

const eventId = lumni.trackAi({ userId: "u1", event: "chat", model: "gpt-4o", input: "hi", output: "hello" });
lumni.trackSignal(eventId, "thumbs_up", { sentiment: "POSITIVE" });

Streaming / Multi-Turn

const interaction = lumni.begin({ userId: "user-123", event: "code_gen", input: "Write fibonacci" });
// ... streaming happens ...
interaction.finish("function fib(n) { ... }");

Tracing (Higher-Order Functions)

// Wrap functions with span tracking
const myFlow = lumni.withInteraction("my_flow", async (input: string) => {
  return await callLLM(input);
});

const search = lumni.withTool("web_search", async (query: string) => {
  return await fetch(`https://api.search.com?q=${query}`).then(r => r.json());
});

Span Context Managers

const result = await lumni.toolSpan("database_query", async (ctx) => {
  ctx.recordInput({ sql: "SELECT * FROM users" });
  const data = await db.execute("SELECT * FROM users");
  ctx.recordOutput(data);
  return data;
});

Lifecycle

await lumni.flush();     // Force send buffered events
await lumni.shutdown();  // Flush + stop background timer

Supported Providers (Auto-Instrumented)

| Provider | Package | Status | |----------|---------|--------| | OpenAI | openai | Supported | | Anthropic | @anthropic-ai/sdk | Supported | | Google Gemini | @google/genai | Supported |

Works with any framework that uses these providers: LangChain.js, Vercel AI SDK, OpenAI Agents SDK, etc.

Development

git clone https://github.com/lumni-ai/lumni-js.git
cd lumni-js
npm install
npm run build

License

MIT