Automated chaos engineering via multi-agent reasoning on Microsoft Foundry IQ.
Built on Foundry IQ via Azure AI Foundry Agent Service. All five reasoning agents are orchestrated through Azure AI Agent Service using GPT-4.1-mini, with full observability through Foundry's agent tracing.
Microsoft Agents League Hackathon — Reasoning Agents track
ChaosProof is an autonomous chaos engineering platform powered by Microsoft Foundry IQ multi-agent reasoning.
Unlike traditional chaos engineering tools (Chaos Monkey, LitmusChaos) that execute predefined failures, ChaosProof reasons about the current state of the system before acting.
The system:
- Reads live telemetry and traces from distributed services
- Builds a service dependency graph and computes blast radius
- Plans safe chaos experiments automatically — aborts if blast radius is too large
- Executes fault injections in a sandboxed Docker environment
- Generates concrete remediation artifacts (circuit breaker configs, k8s patches)
- Validates recovery through a critic loop — retries with feedback if recovery fails
| Agent | Model | Role |
|---|---|---|
| Chaos Planner | gpt-4.1-mini | Reads telemetry, proposes 3 ranked fault experiments |
| Graph Analyst | gpt-4.1-mini | Computes blast radius, ranks experiments safest-first |
| Fault Injector | gpt-4.1-mini | Simulates latency, crash, resource exhaustion in sandbox |
| Remediator | gpt-4.1-mini | Diagnoses root cause, generates fix config |
| Verifier / Critic | gpt-4.1-mini | Checks recovery, triggers retry loop if fail |
Each agent reasons from the previous agent's output — this is a genuine multi-step reasoning chain, not a pipeline.