Detect and diagnose Kubernetes failures automatically using AI trained on cluster logs and events.
Automatically apply safe, reversible fixes like rollbacks or scaling within controlled risk tiers.
Benchmark different AI models on how well they choose the correct remediation action.
Run chaos scenarios to test self-healing behavior on a lab Kubernetes cluster.
| syswe/aura | 2arons/llm-cli | abe238/claude-video-plus | |
|---|---|---|---|
| Stars | 11 | 11 | 11 |
| Language | Python | Python | Python |
| Setup difficulty | hard | easy | easy |
| Complexity | 5/5 | 2/5 | 3/5 |
| Audience | researcher | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires a local Kubernetes cluster, Redis, and either a cloud API key or a local Ollama model.
AURA stands for Autonomous Understanding and Remediation Architecture. It is a research project that helps Kubernetes clusters heal themselves when something breaks, but it keeps a human-controlled safety layer between the AI and the actual cluster. An AI model looks at scattered signals such as container logs, cluster events, and pod status, works out what likely caused a problem, and suggests a fix. A separate, rule-based system then decides whether that suggested fix is ever allowed to run. In a test campaign, the project injected seven different failures plus one harmless traffic spike into a lab cluster. It correctly detected all seven real problems, correctly stayed quiet during the harmless spike, diagnosed the root cause every time, and safely carried out six repairs that were later confirmed to have actually fixed the problem, with no incorrect fixes applied. One case involved a failure outside the systems it was allowed to touch, and instead of acting, it correctly refused and flagged the issue for a person to handle. Safety comes from a tier system. Low-risk actions like scaling a workload run after a test pass. Riskier but reversible actions, such as rolling back a deployment, wait through a cooldown period. Actions needing judgment go to a human approval queue, and destructive actions are always blocked. Extra checks include limiting which parts of the cluster can be touched, escalating caution if too many pods would be affected, and testing every change with a dry run before it is applied for real. The project also compared several AI models on how well they picked the right fix, finding that a smaller model running locally on your own machine performed about as well as a cloud-based model, just more slowly, and that a larger local model actually did worse because it ignored the required output format. AURA is written mainly in Python and requires a local Kubernetes cluster, a few command-line tools, Redis, and either a cloud API key or a local Ollama model to try out. The authors describe it as a research prototype tested on a single lab cluster, with a paper on the work currently under review.
A research prototype that lets an AI diagnose Kubernetes failures and propose fixes, while a separate rule-based safety gate controls which fixes actually reach the cluster.
Mainly Python. The stack also includes Python, Kubernetes, Redis.
Setup difficulty is rated hard, with roughly 1h+ to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.