Langfuse

Overview

Langfuse gives you X-ray vision into your AI agents. Every LLM call, tool use, and agent step is traced, timed, and costed — so you can debug failures, optimize performance, and track costs.

How It Works

1. Instrument — Add tracing to agent code (2 lines)
2. Observe — See every step in a visual timeline
3. Evaluate — Run automated quality evaluations
4. Optimize — Identify slow steps and cost hotspots

Use Cases

• Debugging — Trace failures to exact LLM calls
• Cost tracking — Monitor spend per user/feature
• Quality assurance — Automated output evaluation
• Performance — Identify latency bottlenecks

Getting Started

from langfuse import Langfuse
langfuse = Langfuse()
trace = langfuse.trace(name="research-agent")
span = trace.span(name="web-search")
# ... your logic
span.end()

Example

Dashboard:
├── Research Agent (4.2s, $0.08)
│   ├── Query Planning (0.3s, $0.01)
│   ├── Web Search (1.2s)
│   ├── Synthesis (1.5s, $0.05)
│   └── Report Gen (0.4s, $0.02)
⚠️ Web search returning 0 results 15% of the time

Alternatives

• AgentOps — Agent-specific monitoring
• LangSmith — LangChain’s platform
• OpenTelemetry — Generic observability