Langtrace

Langtrace takes an OpenTelemetry-native approach to LLM observability, which makes it composable with existing observability stacks rather than a fully proprietary island. Auto-instrumentation for major providers (OpenAI, Anthropic) and frameworks (LangChain, LlamaIndex) keeps integration friction low. Good fit for teams that want open standards and self-hosting control. Confidence is docs-derived.

Python and TypeScript SDKs with auto-instrumentation patches. One-line SDK initialization auto-instruments compatible libraries. Manual spans available for custom logic. API for querying traces, sessions, and evaluation datasets. OpenTelemetry-compatible export means traces can feed existing observability infrastructure.

API key auth for cloud or self-hosted. Keys from Langtrace dashboard. HTTPS enforced. Self-hosted deployments manage their own auth. SDK initialization accepts the API key and endpoint. The OpenTelemetry heritage means auth patterns align with established conventions.

SDK failures in trace export are non-blocking — agent code continues even if telemetry upload fails. Sampling and batching reduce overhead in high-throughput scenarios. Self-hosted reliability is the operator's responsibility. Open-source with active maintenance on GitHub.

docs.langtrace.ai covers quickstart, SDK reference, framework integrations, and self-hosting. Getting started is among the fastest in LLM observability tooling. Community via Langtrace Discord and GitHub (open-source).