// the find
Scale3-Labs/langtrace
Langtrace 🔍 is an open-source, Open Telemetry based end-to-end observability tool for LLM applications, providing real-time tracing, evaluations and metrics for popular LLMs, LLM frameworks, vectorDBs and more.. Integrate using Typescript, Python. 🚀💻📊
Langtrace is an OpenTelemetry-based observability platform for LLM applications — traces, costs, latency, and evaluations in one place. It targets teams already building on OpenAI, Anthropic, LangChain, or similar stacks who want visibility into what their agents are actually doing. Self-hostable via Docker Compose, with a managed cloud option.
Genuine OTel compliance means you can pipe traces to any compatible backend (Jaeger, Grafana Tempo, Honeycomb) without lock-in. The integration list is wide and honest about gaps — the table showing what's missing in TypeScript vs Python is more useful than the usual 'all supported' claim. Dual-database architecture (Postgres for metadata, ClickHouse for spans) is the right call at scale — ClickHouse handles append-heavy time-series trace data far better than Postgres would. SDKs are Apache 2.0 licensed even though the server is AGPL, so instrumentation doesn't contaminate your app's license.
The TypeScript SDK has noticeably worse framework coverage than Python — no LangChain, no LangGraph, no LiteLLM, no Ollama. If you're a TypeScript-first shop using LangChain, you're blocked. Last push was November 2025 and the repo shows signs of slowing momentum; 1,200 stars after a few years of active development is modest for a space this crowded. Self-hosting requires running two databases (Postgres + ClickHouse) which raises the ops bar considerably for small teams. The evaluation story — datasets, scoring, human eval — looks bolted on rather than designed in, and the UI screenshots suggest it's not yet production-grade compared to dedicated eval tools like promptfoo or Braintrust.