Vibe Coding Discoveropeninference
View on GitHubOpenTelemetry Instrumentation for AI Observability
OpenInference is an OpenTelemetry-based spec plus Python/JS/Java instrumentation packages that trace LLM and agent calls. It emits standardized spans for models, tools, retrievals, and MCP, exportable to any OTel-compatible backend such as Arize Phoenix.
Use Cases
Trace LLM and agent invocations across frameworksSend spans to any OpenTelemetry backend for AI observabilityDebug multi-agent and tool-calling workflowsMonitor token usage, latency, and retrieval spansInstrument RAG pipelines and vector store callsAdd standardized semantic conventions to AI tracesCapture multimodal (image/audio) span attributesEvaluate agent traces in Arize Phoenix or Arize AX
Built With
- Language
- Python
- Frameworks
- OpenTelemetry · LangChain · LangGraph · LlamaIndex · DSPy · CrewAI · Haystack · OpenAI Agents SDK · Claude Agent SDK · Pydantic AI · Smolagents · Agno · MCP · Guardrails · AWS Bedrock · Vertex AI
Tags
observability · tracing · opentelemetry · llmops · instrumentation · telemetry · evals · monitoring · span-attributes · ai-observability · debugging · aiops · semantic-conventions · agent-tracing · multimodal · python