MobiAgentLab
Recorded Demo · Synthetic Incidents · Deterministic Fixture Model

AI Agent Reliability & Observability Lab

Inspect how an AI agent behaves when systems fail.

Follow queue state, graph steps, retries, failures, and validated diagnoses in one place.

Explore verified scenariosReview evaluation evidence
Recorded demo

Four sanitized runs from the local pipeline. Read-only; no model calls.

Runtime
Recorded
Evidence
Sanitized
Loading recorded evidence

Architecture

Execution path

Run history →Evaluations →
  1. 1

    FastAPI

    accepts allowlisted synthetic incidents

  2. 2

    Redis / RQ

    dispatches work with at-least-once delivery

  3. 3

    LangGraph

    runs bounded investigation steps and retries

  4. 4

    PostgreSQL

    owns lifecycle state and sanitized events