Langtrace

Langtrace is an open-source observability platform purpose-built for monitoring LLM applications and AI agents. Built on the OpenTelemetry standard, Langtrace provides distributed tracing, cost tracking, and performance analytics that give developers complete visibility into how their agents behave in production. The platform captures every LLM call, tool invocation, and chain step with detailed telemetry data.

The SDK integrates with minimal code changes — typically a single initialization line — and automatically instruments popular frameworks including LangChain, LlamaIndex, CrewAI, DSPy, and Anthropic's SDK. This auto-instrumentation captures prompts, completions, token counts, latency, model parameters, and costs without manual logging code.

Langtrace's tracing dashboard shows the complete execution flow of agent requests with waterfall visualizations, making it easy to identify bottlenecks, failed tool calls, and unexpected agent behaviors. Each trace includes detailed information about LLM interactions, retrieval steps, and tool executions, enabling root cause analysis when agents produce incorrect or slow results.

Cost tracking is a standout feature — Langtrace automatically calculates costs for every LLM call based on model pricing, providing per-request, per-user, and per-feature cost breakdowns. This is essential for teams managing agent budgets and optimizing token usage.

The platform supports both self-hosted deployment (via Docker) and a managed cloud service. Self-hosted deployment uses ClickHouse for efficient trace storage and provides full data sovereignty. The evaluation features enable teams to rate agent outputs and build datasets for systematic quality assessment. Langtrace represents the OpenTelemetry-native approach to LLM observability, complementing general APM tools with agent-specific insights.