langprobe
Python SDK

Install

One package, langprobe, with everything as optional extras — install only what you need.

The Python SDK is one package, langprobe, with everything behind optional extras. Install only what you use.

pip install langprobe
InstallWhat you get
langprobeNative SDK — LangprobeClient (.ingest + .control), the @trace decorator, the with span() context manager, identify(). httpx-only, no heavy deps.
langprobe[otel]langprobe.init(...) — an OpenTelemetry tracer provider + OTLP/protobuf exporter to /v1/traces in four lines.
langprobe[crewai] (or dspy, pydantic-ai, openai-agents, llama-index, openai, anthropic, litellm)The matching OpenInference instrumentor, wired by init(instrumentations=[...]).
langprobe[langchain]LangChain / LangGraph callback bridge — LangprobeCallbackHandler.
langprobe[langsmith]LangSmith-compat shim — Client, traceable.

Combine extras as needed — e.g. framework tracing in four lines:

pip install "langprobe[otel,crewai]"

Which one do I want?

  • New code, want traces fast → langprobe[otel] and init().
  • New code, want manual control → langprobe and the native API (@trace / span).
  • A framework (CrewAI, DSPy, Pydantic AI, OpenAI Agents, LlamaIndex) → langprobe[otel,<framework>].
  • Existing LangSmith / LangChain code → the integrations.

Not on Python? You don't need this SDK — langprobe ingests plain OTLP/HTTP, so any language's OpenTelemetry exporter pointed at /v1/traces works. See Tracing & instrumentation.

Self-hosting

Run langprobe in your VPC — one compose file for local, a Helm chart and operator for Kubernetes.

Native usage

LangprobeClient, the @trace decorator, the span context manager, and identify() — no OpenTelemetry required.

On this page

Which one do I want?