About Agentic Stacks

Agentic Stacks packages domain expertise into installable skill packs that teach AI agents how to deploy, manage, and troubleshoot complex infrastructure.

The Problem

Agentic AI is transforming how teams manage infrastructure, but general-purpose AI coding assistants lack the deep, specific knowledge that infrastructure operations demand — the kind that takes DevOps and SRE teams years to build. An agent needs to know not just what commands to run, but when to run them, what order matters, what can go wrong, and how to recover.

This knowledge exists in documentation, runbooks, playbooks, and the heads of senior engineers. It's scattered, version-specific, and hard to keep current. Self-hosted and on-premise infrastructure like OpenStack and Kubernetes is especially demanding — there are no managed services to fall back on. Agentic Stacks makes this operational expertise portable and composable.

What Is a Stack?

A stack is a git repository containing skills — structured markdown files that teach an agent how to operate in a specific domain. Each skill covers a distinct operational concern: cloud deployment, configuration management, troubleshooting storage, incident response procedures, or choosing between networking options.

Unlike infrastructure as code tools that define desired state, stacks teach agents the operational knowledge needed to get there — the deployment order, the safety checks, the troubleshooting steps. They complement your existing IaC and configuration management tooling by giving the agent the context it needs to generate correct configurations and run deployments safely.

Stacks also include a manifest (stack.yaml) for machine-readable metadata and a CLAUDE.md that sets the agent's identity, safety rules, and routing table.

How It Works

# Create a project and pull stacks
agentic-stacks init my-cluster
cd my-cluster
agentic-stacks pull kubernetes-talos

# Add more stacks for cross-domain expertise
agentic-stacks pull hardware-dell

# Open in Claude Code, Codex CLI, Gemini, or any coding agent
# The agent reads .stacks/*/CLAUDE.md and combines expertise
# Output is native format — globals.yml, inventory, configs

When a coding agent enters the project directory, it reads each stack's CLAUDE.md and gains the combined expertise. This AI pair programming workflow means the agent handles the domain-specific details — BIOS configuration, RAID setup, firmware updates, network boot sequences — while you make the high-level decisions. A project with an OpenStack stack and a hardware stack produces an agent that can configure Dell servers and deploy OpenStack on them.

Who Is This For?

Agentic Stacks is AI for developers, sysadmins, and ops teams operating real infrastructure. Whether you're running a homelab, a private cloud, a data center, or a hybrid cloud environment:

  • DevOps engineers deploying and managing Kubernetes, OpenStack, or Docker containers
  • SRE teams codifying incident response procedures and production operations knowledge
  • Sysadmins managing bare metal servers — Dell, HPE, Supermicro — across generations
  • Homelab enthusiasts deploying their first Kubernetes cluster or Ceph storage pool
  • Platform teams building deployment automation and operational tooling

Works with Any Agent

Stacks are plain markdown — they work with any LLM tools and AI coding assistants. Claude Code reads CLAUDE.md automatically. OpenAI Codex CLI reads AGENTS.md. Gemini CLI reads GEMINI.md. Editors like Cursor, Windsurf, VS Code with GitHub Copilot, and JetBrains with AI plugins can all reference stack skills. Models from Anthropic, OpenAI, Google, and open source AI projects all work because the knowledge is structured text, not model-specific.

Composition

Stacks are designed to compose. A hardware stack handles bare metal provisioning — BIOS configuration, RAID setup, firmware updates, and network boot. A platform stack handles Kubernetes deployment or OpenStack deployment. A storage stack handles Ceph storage or NFS. Pull what you need for your on-premise or hybrid cloud deployment — the agent merges the expertise.

The Feedback Loop

Stacks get smarter over time. As you work with an agent, ask it to capture notes when things go wrong — a command that fails, a config that needs a workaround, a version-specific gotcha. Site reliability teams can feed incident response learnings back into the stack as known issues, updated procedures, and better decision guides.

Tip: ask your agent to take notes as you go

"Hey, that NTP fix we just did — add it to known issues
 for this version so the next person doesn't hit it."

Every operator encounter is a chance to improve the stack. File an issue, open a PR, or just tell your agent to document what happened. The next person who pulls the stack gets the benefit of your production operations experience.

Get Involved

Agentic Stacks is open source (MIT license). Whether you're exploring AI DevOps workflows or building internal tooling, authoring a stack is a way to share what your team has learned. Stacks work with open source AI models too — any agent that reads markdown can leverage the skills, regardless of the underlying model. Browse existing stacks to see what's available, or check the FAQ to get started.