Free Alternative to LangSmith

Langfuse is the closest free alternative to LangSmith for real LLM application observability. The MIT-licensed core is fully self-hostable, so your traces stay on your infrastructure, and the cloud Hobby tier starts free with a usable monthly unit allowance. In August 2026 it remains the community default when teams leave LangSmith’s $39/seat Plus plan—especially after Langfuse’s ClickHouse acquisition, which has so far kept the open core intact. On a Mac you typically run the self-hosted stack via Docker or a remote Linux box while developing locally with LangChain, LangGraph, or OpenTelemetry instrumentation. You get traces, sessions, prompt management, datasets, and evals without paying per seat. The honest gap versus LangSmith: LangChain-native polish, one-click hosted agent deployment (LangSmith Deployment/Fleet/Engine), and enterprise support still favor LangSmith. Self-hosting also means you own Postgres/ClickHouse ops, backups, and upgrades. If you only need Claude Code session forensics rather than production agent tracing, pair Langfuse with Claude DevTools instead of forcing one tool to do both jobs.

Claude DevTools is a free, open-source desktop application that reconstructs exactly what Claude Code did during your sessions. Unlike general-purpose observability platforms, it focuses specifically on visualizing Claude Code's hidden execution data—parsing the raw session logs already stored on your machine and presenting them in an intuitive, browsable interface. I re-tested it on my M2 MacBook Pro in August 2026 and found it remarkably effective for understanding what actually happened during complex coding sessions. The app displays every file touched, every tool call made, token usage breakdowns, subagent invocations, and context window utilization—all without requiring any integration or code changes. It does not run or modify Claude Code itself; it simply reads and visualizes what the CLI already recorded. For developers who primarily use Claude Code and want to understand why certain decisions were made or trace the evolution of a multi-file edit, this is an essential tool. The visual timeline makes it trivial to spot when Claude went down a wrong path or exceeded token limits. However, it is purpose-built for Claude Code—if you need observability for custom LangChain agents, OpenAI API calls, or other LLM frameworks, you will need different tooling.