Non-Invasive AI Agent Observability With OBI
Talkat KubeCon + CloudNativeCon + OpenInfra Summit + PyTorch Conference China 2026 · 2026-09-07..09-09·
summary
OBI uses eBPF kernel probes to intercept network traffic and extract plain‑text MCP JSON‑RPC, MCP Streamable HTTP, embedding and reranking calls, and tool invocations from OpenAI, Anthropic, Gemini and Qwen without changing application code. A single daemonset process per node parses protocols, emits OpenTelemetry GenAI semantic spans and propagates trace context across Go, Python and NodeJS services. Runs in production at hundreds of nodes and thousands of microservices with minimal overhead, finds issues like a 320% token spike and broken RAG steps, enabling zero-instrumentation observability and consistent cross‑provider telemetry.
AI-written summary · about
chapters (9)
- 0:00AI agent observability intro
- 1:05OpenTelemetry BPF instrumentation
- 4:56OBI vs traditional OpenTelemetry
- 15:38Why AI agents need observability
- 16:46OBI as APM complement
- 19:36Four layers of AI agent observability
- 21:28OBI case studies
- 24:47OBI future roadmap
- 25:55Q&A
related talks
20mGenAIOps - OpenSearch for AI & GenAI Platform Observability
31mDeep Dive To AI Agent Observability
23mObserving the Observers: Bringing OpenTelemetry to Autonomous AI Agents
38mAI Monitoring Mastery: Optimize Performance, Ensure Compliance, and Cut Costs
27mObservability as a foundation for AI-enabled platforms
40mOptimizing AI Applications for Production with Observability
22mAI Observability At Meta Scale- 27mObservability as a foundation for AI-enabled platforms