Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for KubeIntellect: Qwen-Powered Kubernetes Autopilot Agent
Inspiration
Kubernetes is powerful, but operating it under pressure is still difficult. When a service fails, engineers often need to jump between kubectl, logs, metrics, events, dashboards, and previous incident context before they can understand the root cause. This slows down incident response and creates risk, especially when remediation actions are executed manually during stressful moments.
KubeIntellect was created to act as an AI DevOps engineer for Kubernetes: an agent that can investigate incidents, explain what is happening, and prepare safe remediation steps while keeping humans in control.
What it does
KubeIntellect is a Qwen Cloud-powered Kubernetes Autopilot Agent. Users can ask natural-language questions such as:
- “Why is my pod crashlooping?”
- “Show recent errors for this service.”
- “Investigate why latency increased.”
- “Prepare a safe rollback plan.”
The agent connects to Kubernetes, Prometheus, Loki, and cluster events to inspect the system from multiple angles. For simple questions, it can directly query the right tool. For complex incidents, it decomposes the task across specialized subagents for pods, metrics, logs, and events, then combines the evidence into a clear root-cause analysis.
High-risk or destructive operations are not executed automatically. They are routed through a human-in-the-loop approval gate so the operator can review the proposed command before execution.
How we built it
KubeIntellect uses a FastAPI backend with an agent orchestration layer. The coordinator agent interprets the user request, selects tools, delegates investigation tasks, and synthesizes the final response. Specialized subagents inspect Kubernetes resources, metrics, logs, and events.
For the hackathon version, the reasoning layer is being adapted to use Qwen models on Qwen Cloud, with the backend deployed on Alibaba Cloud. The system also includes a CLI client called kube-q, role-based access control, persistent state, audit logging, and deployment options for local, Docker Compose, and Kubernetes environments.
Challenges we ran into
The biggest challenge is balancing autonomy with safety. A Kubernetes agent should be useful enough to diagnose and prepare remediation, but it must not blindly execute risky commands. We designed the system around permission levels, approval gates, and auditability so the agent can help operators without removing human responsibility.
Another challenge is combining different operational data sources. Kubernetes state, events, metrics, and logs all tell different parts of the story. The multi-agent design helps separate these concerns and makes the final diagnosis more reliable.
What we learned
We learned that production-ready AI agents need more than a strong model. They need clear tool boundaries, permission checks, observability, persistent state, and safe failure modes. In DevOps workflows, the agent’s reasoning is only one part of the system; the surrounding engineering architecture is what makes it trustworthy.
What’s next
Next, we plan to improve Qwen Cloud integration, expand the remediation playbook library, add stronger evaluation benchmarks for incident diagnosis, improve memory across repeated incidents, and support more Kubernetes production environments.
Built With
- access
- alibaba-cloud
- docker
- docker-compose
- fastapi
- helm
- human-in-the-loop-approval
- kind
- kube-q
- kubectl
- kubernetes
- langchain
- langfuse
- langgraph
- logql
- loki
- postgresql
- prometheus
- promql
- python
- qwen-cloud
- rest-api
- role-based
- sqlite
Log in or sign up for Devpost to join the conversation.