▸ Open source and self-hostable.
Open-source LLM engineering platform: tracing, evals, prompts.
Trace, debug and evaluate agents
Quick answer: Best overall to trace, debug and evaluate agents: Langfuse. Open source and self-hostable. Runner-up: LangSmith (Native for LangChain/LangGraph, works framework-agnostic). Also strong: Braintrust. We list 8 tools below, ranked by quality and fit.
Observability, evals and cost tracking for LLM/agent apps.
Top picks, ranked
▸ Native for LangChain/LangGraph, works framework-agnostic.
Tracing, evals and monitoring for LLM apps and agents.
▸ Strong eval-driven development workflow.
Evals and observability platform for AI products.
▸ OpenTelemetry-based tracing and evals.
Open-source AI observability and evaluation.
▸ One-line proxy for cost and latency tracking.
LLM observability, routing and AI gateway.
▸ Agent-specific session replays.
Observability and session replay for AI agents.
▸ Cuts agent spend in-flight (routing, cache, loop guards) with a free Watchtower tier.
Cost-control layer for AI agents: routes, caches, and trims spend while keeping the same output.
▸ Policy rules plus human review for consequential agent actions, with full audit trail.
Open-source human approval layer that intercepts agent tool calls and routes risky ones to people.
Related use cases
Run agent-generated code in a sandboxBuild an AI agent in codeBuild agent workflows without codingLaunch a phone/voice AI agentLet an agent use a web browserGive your agent live web search and data
More in AI Agents.