LangSmith, a prominent platform for AI application monitoring, has announced its integration with OpenTelemetry, enhancing its capabilities for distributed tracing and observability, according to LangChain. This integration allows LangSmith to ingest traces in the OpenTelemetry format, providing developers with a comprehensive view of their application’s performance.

OpenTelemetry Integration Details

OpenTelemetry is an open standard for distributed tracing and observability that supports a wide range of programming languages, frameworks, and monitoring tools. This integration means that LangSmith’s API layer can now directly accept OpenTelemetry traces. Developers can point any supported OpenTelemetry exporter to the LangSmith OTEL endpoint, ensuring their traces are ingested and accessible within LangSmith. This setup provides a unified view of application performance, combining LLM monitoring with system telemetry.

Semantic Conventions and Supported Formats

OpenTelemetry defines semantic conventions for various use cases, including databases, messaging systems, and protocols such as HTTP or gRPC. LangSmith is particularly focused on conventions for generative AI, a developing area with few existing standards. Currently, LangSmith supports traces in the OpenLLMetry format, which facilitates out-of-the-box instrumentation for different LLM models, vector databases, and common frameworks. Future plans include support for other semantic conventions as they evolve.

Getting Started with OpenTelemetry

To utilize this new feature, developers can start with an OpenTelemetry-based client, such as the OpenTelemetry Python client. By installing necessary dependencies and configuring environment variables, developers can begin tracing their applications. The LangSmith dashboard will display these traces, providing insights into application performance.

Additional SDK Integrations

LangSmith also supports integrations with other SDKs, such as Traceloop and Vercel AI SDK. These integrations allow developers to send tracing data using various SDKs, offering flexibility and compatibility with different AI models and frameworks. For instance, the Traceloop SDK supports a broad range of integrations, and the Vercel AI SDK offers a client-side trace exporter defined by the LangSmith library.

These advancements position LangSmith as a robust solution for developers seeking comprehensive observability and performance monitoring in AI applications, leveraging the capabilities of OpenTelemetry to provide a detailed and integrated view of system operations.

Image source: Shutterstock