William-Lu-stack/Flawless

What it solves

Flawless is an AI-native SRE (Site Reliability Engineering) control plane designed to move beyond simple AI chat interfaces for infrastructure management. It solves the problem of fragmented observability and risky, unverified manual fixes in Kubernetes and cloud environments by connecting discovery, diagnosis, and remediation into a single, auditable loop.

How it works

Flawless implements an "AgenticOps Loop" (discover → diagnose → preview → approve → execute → verify → learn). It uses a Large Language Model (LLM) as a planner and explainer, but keeps the actual execution boundary separate. The system collects evidence (logs, metrics, topology, and events), generates a remediation plan, requires human approval, executes the change via controlled tools (like Argo Rollouts for progressive delivery), and then verifies that the original symptom is actually gone rather than just assuming a successful command execution.

Who it’s for

It is built for SREs and platform engineers managing Kubernetes and cloud infrastructure who need a safe, automated way to handle incidents, perform root-cause analysis, and maintain a library of reusable operational "Skills."

Highlights

• Verified Recovery: Tests the original symptom after a fix is applied to prove the system has actually recovered.
• Controlled Remediation: Integrates human approval, RBAC, and dry-runs to ensure AI doesn't make unguided changes.
• Progressive Delivery: Uses Argo Rollouts for canary deployments and risk-gated capacity recovery during fixes.
• Skills Library: Encodes expert operational knowledge into portable, reusable skills for agents.
• Topology Impact Analysis: Provides 2D/3D topology and eBPF-based data-flow adapters to analyze the blast radius of failures.
• Knowledge Base: Supports RAG (Retrieval-Augmented Generation) by allowing users to upload runbooks, logs, and documentation.