Lantern — Agent Observability & Tracing

Full-stack visibility into AI agent behavior. The Datadog for AI agents.

The Problem

AI agents make opaque decisions across multiple reasoning steps, tool calls, and model invocations. When an agent produces a bad output, there's no way to trace why. When costs spike, there's no attribution. When quality drifts, nobody notices until users complain. Lantern captures every step of agent reasoning, scores quality continuously, and attributes costs to the token level.

Install

npm install @openlantern-ai/sdk

Quickstart

import Anthropic from "@anthropic-ai/sdk";
import { LanternTracer, ConsoleExporter, wrapAnthropicClient } from "@openlantern-ai/sdk";

// 1. Create tracer
const tracer = new LanternTracer({
  serviceName: "my-agent",
  exporter: new ConsoleExporter(),
});

// 2. Wrap your AI client
const client = wrapAnthropicClient(new Anthropic(), tracer);

// 3. Use as normal — all calls are traced
const res = await client.messages.create({
  model: "claude-sonnet-4-5-20251001",
  max_tokens: 1024,
  messages: [{ role: "user", content: "Hello!" }],
});

await tracer.shutdown();

Live Demo

Try the interactive dashboard locally — no API keys required:

git clone https://github.com/your-org/lantern.git
cd lantern
pnpm install
pnpm demo

This starts an ingest server on http://localhost:4100, seeds it with realistic agent traces from multiple services, and opens the dashboard in your browser. Explore:

• Traces — Click through agent reasoning chains with full input/output visibility
• Metrics — Cost attribution by agent and source, token usage, trace timelines
• Sources — See which services are sending traces, their SDK versions, and connected agents

Supported Integrations

LLM Providers

Provider	Language	Method
Anthropic	TS, Python	SDK wrapper (wrapAnthropicClient)
OpenAI	TS, Python	SDK wrapper (wrapOpenAIClient)
Google Gemini	TS	SDK wrapper (wrapGoogleGenerativeModel)
Mistral	TS	SDK wrapper (wrapMistralClient)
Cohere	TS	SDK wrapper (wrapCohereClient)
AWS Bedrock	TS	SDK wrapper (wrapBedrockClient)
Groq	TS	OpenAI-compatible (wrapOpenAICompatClient)
Together AI	TS	OpenAI-compatible (wrapOpenAICompatClient)
Fireworks AI	TS	OpenAI-compatible (wrapOpenAICompatClient)
DeepSeek	TS	OpenAI-compatible (wrapOpenAICompatClient)
Perplexity	TS	OpenAI-compatible (wrapOpenAICompatClient)
Ollama	TS	OpenAI-compatible (wrapOpenAICompatClient)
OpenRouter	TS	OpenAI-compatible (wrapOpenAICompatClient)
xAI (Grok)	TS	OpenAI-compatible (wrapOpenAICompatClient)
Cerebras	TS	OpenAI-compatible (wrapOpenAICompatClient)
Novita AI	TS	OpenAI-compatible (wrapOpenAICompatClient)

Frameworks & Agent SDKs

Framework	Language	Method
LangChain	TS	Callback handler (createLanternCallbackHandler)
LlamaIndex	TS	Event handler (createLanternEventHandler)
Vercel AI SDK	TS	Function wrapper (wrapGenerateText / wrapStreamText)
OpenAI Agents SDK	TS	Trace processor (createLanternTraceProcessor)
Mastra	TS	Telemetry hook (createLanternMastraHook)
MCP	TS	Tool call wrapper (wrapMcpClient)
CrewAI	Python	Lifecycle handler (create_lantern_crewai_handler)
Pydantic AI	Python	Lifecycle handler (create_lantern_pydantic_handler)
AutoGen	Python	Message hooks (create_lantern_autogen_handler)
Haystack	Python	Pipeline callbacks (create_lantern_haystack_handler)
DSPy	Python	Module tracing (create_lantern_dspy_handler)
Smolagents	Python	Step callbacks (create_lantern_smolagents_handler)

See docs/integrations/overview.md for detailed per-integration documentation.

Core Concepts

• Traces — One complete agent execution, from user request to final response.
• Spans — Individual reasoning steps: LLM calls, tool invocations, retrieval, custom steps.
• Scorers — Automated quality checks: relevance, toxicity, latency, or your own custom scorer.
• Baselines — Score snapshots for regression detection. Get alerted when quality drops.
• Cost attribution — Token-level cost tracking by model, agent, and workflow.

OSS vs Enterprise

Feature	OSS (MIT)	Enterprise (BUSL-1.1)
Full trace capture (SDK)	✓	✓
SQLite + Postgres storage	✓	✓
Dashboard (traces, metrics)	✓	✓
Latency + cost scorers	✓	✓
Custom eval scorers	✓	✓
Self-hosted deployment	✓	✓
PII detection in traces	—	✓
SOC2/HIPAA audit export	—	✓
Slack/PD/webhook alerts	—	✓
Team-scoped RBAC	—	✓
Managed cloud ingest	—	✓

Self-Hosting

docker compose -f docker/docker-compose.yml up -d

See docs/self-hosting.md for details.

Contributing

1. Fork the repository
2. Create a feature branch
3. Run pnpm build && pnpm test && pnpm typecheck before submitting a PR
4. Open a pull request against main

See ARCHITECTURE.md for architecture and conventions.

License

MIT — see LICENSE. Enterprise features are licensed under BUSL-1.1.