50 ideas · architecture documented multi-platform human-in-the-loop

Where AI actually meets DevOps.

An open catalog of 50 practical implementation ideas for applying AI to CI/CD, Kubernetes, Infrastructure as Code, Cloud, Observability, Security and Platform Engineering — each with a real problem, a proposed solution and a documented architecture. Provider-agnostic. Security-first. Human-in-the-loop.

Browse the 50 ideas Read on GitHub
50
ideas documented, each with an architecture
10+
focus areas across the DevOps landscape
40+
AI platforms & models referenced with roles
100%
problem-first · human-in-the-loop · security-first

Why this catalog

Not “add AI everywhere”.
Add it where it earns its keep.

Every idea starts from a real DevOps pain point and works forward to an architecture — never the other way around.

Problem-first

Each idea opens with the concrete toil or risk it removes — log archaeology, flaky triage, IAM sprawl — before any model is mentioned.

Architecture documented

Components, data flow, diagrams and platform mapping for all 50 ideas — enough to start a prototype tomorrow.

Multi-platform by design

Frontier, efficient, open-weight and local models — with gateways for routing, fallbacks and budgets. Swap any layer.

Human-in-the-loop

Explicit approval tiers everywhere: the AI investigates and suggests, humans review and act. Automation is earned, not assumed.

Security first

Redaction before model calls, least-privilege tool access, immutable audit trails and prompt-injection defenses.

Cost-aware

Event-driven workloads, cheap model tiers for triage, frontier models for deep analysis, and local-model escape hatches.

Living demos

Watch an idea run,
start to finish.

Animated walk-throughs of three ideas from the catalog — the same investigate → explain → draft-a-fix shape every idea follows. They loop like gifs; switch tabs to change the story.

The AI platform landscape

Pick per layer,
not per vendor.

Ideas reference platforms by role — reasoning, routing, retrieval, tool access, observability — so any layer can be swapped without redesigning the rest.

Frontier models (cloud)
  • OpenAI GPT-5.x
  • Claude Opus/Sonnet 4.x
  • Gemini 3 Pro
  • xAI Grok
  • Z.ai GLM
Cost-efficient tiers
  • GPT-5 Mini/Nano
  • Claude Haiku 4.5
  • Gemini Flash
  • DeepSeek V4
  • Mistral Small
Routing & gateways
  • LiteLLM
  • OpenRouter
  • Portkey
  • Kong AI Gateway
  • Cloudflare AI Gateway
Open-weight / local
  • DeepSeek V4
  • Qwen 3.x
  • Llama 4
  • Mistral Large 2
  • via vLLM · Ollama
Agent frameworks
  • LangGraph
  • Microsoft Agent Framework
  • OpenAI Agents SDK
  • Google ADK
  • PydanticAI
  • CrewAI
MCP servers (DevOps)
  • GitHub
  • kubernetes-mcp-server
  • mcp-grafana
  • Terraform
  • AWS / Azure
  • PagerDuty
Kubernetes AI (CNCF)
  • K8sGPT
  • HolmesGPT
  • kagent
  • kubectl-ai
AI observability & evals
  • Langfuse
  • Arize Phoenix
  • LangSmith
  • W&B Weave
  • OTel GenAI
  • promptfoo
Code & PR reviewers
  • CodeRabbit
  • Greptile
  • Qodo
  • Graphite
  • GitLab Duo
RAG stack
  • pgvector
  • Qdrant
  • Weaviate
  • OpenSearch k-NN
  • BGE-M3 embeddings

The idea atlas

All 50 ideas.
One screen.

Every cell is one idea, colored by focus area. Hover for the name, click to open the full brief — a fast way to scan the landscape before diving into the catalog below.

The catalog

50 ideas.
One of them is your next weekend build.

Filter by focus area and complexity, search across names, problems and integrations, or roll the dice with $ random. Click any idea for the full brief — problem, workflow, architecture and platforms.

Open source · MIT · ideas → architectures → implementations

Ship the first implementation.

Phase 1 — the ideas — is documented. Phase 3 is yours: pick an idea, prototype it against free tiers or local models, and contribute it back. The catalog grows by being built.

Browse ideas Read CONTRIBUTING.md