From Tool Calls To Context Fabric: Building AI-Native Observability for…
summary
An AI-native observability fabric precomputes context graphs and metadata caches so agents query Model Context Protocol (MCP) endpoints instead of scanning raw stores, linking metrics, logs, anomalies and docs via knowledge graphs and caches in DuckDB/Redis with Apache Jena/AWS Neptune for relationships and Apache Ranger for access control. Operates at 3 trillion datapoints and 2 trillion active series per day across Thanos/Prometheus/Loki, detects a 6x query fan‑out causing retries (P95 47s vs baseline 5s), and the claim‑gate confirms evidence without auto-remediation.
AI-written summary · about
chapters (8)
- 0:00AI-native observability intro
- 2:35Tool-based AI agents limits
- 8:23Shared context for alerts
- 15:02Live evidence demo
- 21:30MCP observability architecture
- 24:17Context layers from time-series
- 27:29Reusable investigation workflows
- 28:55Closing and open-source
related talks
24mBuilding Queryable O11y with the Model Context Protocol
29mBig Graphs, Tiny Contexts: Dev Tools for Agents
27mAI-Powered Root Cause Analysis at Scale: From Theory To Production...
28mBeyond Dashboards: Architecting AI Agents for Autonomous Observability
37mContext is King in AI Agents
20mFrom Data Dumps To Smart Context: Building MCP Servers That AI Can Actually Use
29mNon-Invasive AI Agent Observability With OBI
05mThe Role of Context in Observability