Chaos Engineering Platform

Part of the worlds-biggest-software-project initiative.

An AI-native, open-source chaos engineering platform for fault injection, blast-radius analysis, and forward-looking resilience scoring across cloud, Kubernetes, and AI-native systems.

The Chaos Engineering Platform helps SRE teams, platform engineers, and engineering leaders validate the resilience of distributed systems through hypothesis-driven failure experiments. It combines a multi-infrastructure fault injection engine with predictive blast-radius modelling and AI-generated experiment hypotheses, targeting both conventional cloud workloads and the new failure modes introduced by LLM-based components.

Why Chaos Engineering Platform?

Commercial leaders are expensive. Gremlin typically costs $20,000–$100,000+/year, putting mature chaos engineering out of reach for mid-market and startup teams.
Cloud-native services are siloed. AWS Fault Injection Service and Azure Chaos Studio offer low-overhead managed chaos but lock teams into a single cloud and provide smaller fault libraries than Gremlin or Steadybit.
Open-source tooling demands heavy lifting. LitmusChaos, Chaos Toolkit, and Pumba are free and flexible, but require significant self-service engineering effort, with maturing UIs and limited blast-radius prediction.
Existing platforms have no AI-native coverage. No incumbent uses LLMs to propose experiments, predict blast radius, or test failure modes specific to LLM-based components, vector DBs, and RAG pipelines.
Resilience scoring is becoming the executive KPI. Forward-looking reliability scores based on active failure testing are replacing lagging uptime metrics — but this capability is concentrated in the most expensive commercial tiers.

Key Features

Fault Injection & Experiment Orchestration

Fault injection across 10–15 core failure types: pod, container, and instance termination; network latency and packet loss; CPU, memory, and disk I/O stress
Coverage spanning Kubernetes plus at least one major cloud (AWS or Azure) at MVP, with an extension path to multi-cloud
Sequential and parallel fault execution within a single scenario to mimic real-world cascading failures
Declarative experiment definition in YAML/JSON for GitOps workflows and version control

Safety, Targeting & Governance

Tag-based and resource-based blast-radius controls to scope experiments precisely
Stop conditions tied to monitoring thresholds (e.g. CloudWatch Alarms, Prometheus queries) to halt runaway experiments
Steady-state hypothesis validation via HTTP, command execution, and custom metric probes
Role-based access control (RBAC) for team usage and compliance audit logging

Observability & Reporting

Real-time monitoring integration with Prometheus and CloudWatch during experiments
Experiment result reporting with pass/fail verdicts, metric comparisons, and audit logs
Resilience scoring as a forward-looking reliability KPI tracked across deployments

Developer & Platform Workflows

CI/CD integration enabling chaos experiments as deployment gates or post-deploy validation
Web UI for visual, timeline-based experiment design alongside YAML authoring
Pre-built experiment library (Chaos Hub equivalent) covering common failure scenarios
Dependency discovery and topology mapping to surface hidden service relationships

Extensibility

Open extension framework for custom faults, probes, and integrations
Pluggable driver architecture for new infrastructure targets without vendor lock-in

AI-Native Advantage

The platform uses AI to close gaps that no incumbent currently addresses: analysing system architecture, dependency graphs, and historical incident data to automatically propose ranked chaos experiments; modelling downstream dependencies to predict blast radius before execution; running an autonomous game day orchestrator that plans, executes, and interprets full scenarios end-to-end with real-time scope adjustment; and providing purpose-built experiments for AI-native systems — prompt injection cascades, context-window exhaustion, and tool-call error propagation in LLM-based components.

Tech Stack & Deployment

The platform is designed for self-hosted deployment on Kubernetes with declarative experiment definitions managed as code. It aligns with the Principles of Chaos Engineering (chaosengineering.io), supports GameDays as a first-class organisational practice, and integrates with SRE error-budget and SLO workflows. Architectural inspiration is drawn from CNCF chaos engineering projects (notably LitmusChaos, which uses a Kubernetes operator pattern with ChaosEngine, ChaosExperiment, and ChaosResult custom resources). Integrations target Prometheus, CloudWatch, and standard CI/CD tooling.

Market Context

The chaos engineering tools market was valued at approximately $2.37 billion in 2025 and is projected to reach $2.56 billion in 2026, with long-range forecasts of 8–24% CAGR through 2030–2033 (Cognitive Market Research; MarketsandMarkets, 2026). Incumbent pricing ranges from custom enterprise contracts at $20,000–$100,000+/year (Gremlin) to pay-per-action-minute cloud services (AWS FIS, Azure Chaos Studio) and free open-source tools that require significant engineering investment. Primary buyers are SRE teams at high-availability services (fintech, eCommerce, streaming), platform engineers embedding chaos into CI/CD, and CTOs in regulated industries demonstrating operational resilience for compliance audits.

Project Status

This project is in the research and specification phase.
Contributions, feedback, and domain expertise are welcome.

Contributing

We welcome contributions from developers, domain experts, and potential users. See CONTRIBUTING.md for guidelines.

Important: All contributions must be your own original work or clearly attributed open-source material with a compatible licence. Copyright infringement and licence violations will not be tolerated and will result in immediate removal of the offending contribution. If you are unsure whether a piece of code, text, or other material is safe to contribute, open an issue and ask before submitting.

Licence

Licence to be determined. See discussion for context.