Guides
Step-by-step guides covering all major aspects of Tumult.
| Guide | Description |
|---|---|
| Experiment Format | TOON experiment structure, all fields and provider types |
| Execution Flow | Five-phase lifecycle, orchestration internals |
| CLI Reference | All commands: run, validate, analyze, export, compliance |
| Statistical Baselines | Data-derived tolerance methods: percentile, IQR, mean/stddev |
| Analytics Guide | DuckDB SQL queries over experiment journals, Parquet export |
| Observability Setup | OTel env vars, collector configs, Jaeger, Grafana, SigNoz |
| Load Testing Guide | k6 and JMeter integration with chaos experiments |
| MCP Guide | The 26-tool MCP server: annotations, structured output, tumult:// resources, and the closed run→ingest→recommend loop |
| ChaosGraph | The typed chaos knowledge graph served to agents over MCP: node/edge model, the two query tools, and bounded, token-efficient agent context (reproducible via make demo-proof) |
| Agentic Quickstart | Fault injection for AI agents: scenario packs, contracts, replay |
| Agentic Live Clients | Inject faults into Claude Code, Codex, OpenCode, and Copilot traffic |
| Agentic Cross-Client Observability | Normalize agent telemetry onto one schema; two-sided spans and trace-nesting tiers per client |
| Agentic Recommendations | Enhance tumult recommend with a local agent CLI (Claude Code, Codex); generate validated experiments |
| Windows Faults | tumult-windows: native process-kill, CPU-stress, and firewall-blackhole faults, validated live against a Windows 11 guest |
Table of contents
- Experiment Format
- Execution Flow
- CLI Reference
- Statistical Baselines
- Analytics Guide
- Observability Setup
- Load Testing Guide
- Agentic Fault Injection Quickstart
- Agentic Fault Injection Against Live Clients
- Agentic Observability
- Token Efficiency
- Agentic Cross-Client Observability
- Agentic Scenarios
- Experiment Scheduling
- Agentic Recommendations
- MCP Guide
- ChaosGraph
- Production Deployment
- Windows Faults