traceloop/openllmetry — reverse-engineered prompt

Build me an open source Python toolkit that helps people see what’s happening inside their LLM app without changing much code. I want developers to install an SDK, add one simple init line, and then automatically get OpenTelemetry tracing for normal app work plus calls to LLM providers like OpenAI, Anthropic, Gemini, Bedrock, HuggingFace, Mistral, Ollama, and vector databases like Pinecone, Chroma, Qdrant, Weaviate, Milvus, and LanceDB.

It should work with existing observability tools instead of locking users into one backend. Make it easy to export to places like Datadog, Honeycomb, Grafana, New Relic, Sentry, Google Cloud, Azure, and the OpenTelemetry Collector.

Please include a clean SDK, separate instrumentations that can be used directly, good getting started docs, examples, tests, and packaging so people can install it with pip. Keep the setup really simple, but make it flexible for teams that already use OpenTelemetry. Look up current OpenTelemetry docs if you need to.