Skip to content

Agentic Investigations

NeuBird doesn't just answer questions — it investigates. Each question triggers an autonomous investigation loop where NeuBird explores your telemetry schema, queries data across multiple sources, correlates the results, and iterates until it reaches a confident conclusion.

How it works

  1. Schema discovery — NeuBird searches your telemetry inventory for relevant data sources using pattern matching across all schemas.
  2. Query execution — It queries your telemetry data through a unified access layer, often running multiple queries in parallel across different data sources.
  3. Evidence correlation — Results from different data sources (incidents, metrics, logs, traces) are correlated to build a complete picture.
  4. Iterative refinement — If initial queries don't provide enough information, NeuBird automatically explores further — checking related tables, adjusting time windows, or drilling into specific services.
  5. Synthesis — Once enough evidence is gathered, NeuBird produces a structured analysis with findings, evidence citations, and recommended actions.

Real-time visibility

Every step of the investigation is visible in the terminal:

  • Tool calls show which tables are being queried and why
  • Progress bar tracks the investigation phase (exploring → investigating → analyzing → synthesizing)
  • Confidence indicator shows how certain NeuBird is about its current hypothesis

Press Ctrl+O to expand any tool call and inspect the raw query and results. See Tool Review for details.

Example questions

  • "What services have the highest error rates right now?"
  • "Why did latency spike on api-gateway at 2am?"
  • "What changed in the last 24 hours that could explain the increase in 503s?"
  • "Which deployments correlate with the performance degradation?"