openinference
by Arize-ai
OpenTelemetry instrumentation for LLMs and agent observability
Summary
Instrument AI applications with OpenTelemetry to capture request traces, metrics, and logs specific to LLM and agent workflows. Provides reusable instrumentation for popular agent and retrieval frameworks so observability is embedded at the SDK level. Includes semantic attributes and spans tailored to LLM calls, embeddings, and retrievals to make agent interactions queryable and linkable across services. See the Human-in-the-Loop pattern for governance considerations and the Agent Registry Pattern to discover and organize tools.
Key Benefits
Ideal For
Teams instrumenting multi-agent or LLM-based systems that need structured traces, metrics, and logs for monitoring and post-hoc evaluation. Consider adopting the Agent Service Mesh Pattern to manage complex tool interactions and security.
Real-World Examples
- Trace and correlate LLM calls across microservices to diagnose agent failures
- Emit standardized metrics and spans for agent delegation and decision points
- Capture retrieval and embedding events for debugging retrieval-augmented agents
- Feed structured telemetry into continuous agent evaluation and governance pipelines