Skip to content

Agentic Ops Framework (AOF)

Overview

The Agentic Ops Framework (AOF) is an open source, Rust-based framework that lets DevOps, SRE, and platform engineers build and orchestrate AI agents using Kubernetes-style YAML specifications and a kubectl-style CLI (aofctl). It is positioned as "n8n for Agentic Ops" — a way to define and run operational AI agents without writing Python, by reusing the declarative mental model operators already have from Kubernetes.

Key Concepts / Architecture

AOF defines three core resource abstractions, each expressed as a Kubernetes-style manifest (apiVersion: aof.dev/v1, kind: ...):

  • Agent — a single AI instance configured with instructions, tools, and a model
  • AgentFleet — a coordinated team of multiple agents working together on a task
  • AgentFlow — a DAG-based, multi-step workflow with conditional branches and human-approval gates

The CLI mirrors kubectl semantics (e.g., aofctl apply, aofctl run agent), so operators familiar with Kubernetes tooling can apply the same workflow to agent definitions.

Key Features

  • LLM providers: OpenAI (GPT-4), Anthropic Claude, Google Gemini, Ollama (local models), and Groq, switchable without code changes
  • Integrations: kubectl, shell/bash execution, HTTP requests, Slack, Discord, Telegram, GitHub, PagerDuty (incident webhooks), and custom MCP tool servers
  • Human-in-the-loop: approval gates implemented as conditional workflow branches — e.g., routing high-severity findings to Slack for manual sign-off before remediation runs
  • Safety controls: allowed command lists and audit logging to constrain what agents can execute against live infrastructure
  • Memory: persistent memory and RAG integration intended to let agents retain context and learn across interactions

Suitable For (Pros)

  • Teams that want operational/DevOps automation agents (incident response, ChatOps bots, remediation workflows) defined declaratively, alongside existing Kubernetes manifests
  • Organizations standardizing on a single CLI/YAML mental model across infrastructure and agent definitions
  • Use cases requiring tight integration with ops tooling (PagerDuty, Slack, GitHub) out of the box

Status

  • Apache 2.0 licensed, written in Rust (~90% of codebase)
  • Beta maturity (latest release v0.4.0-beta at time of writing); installable via Cargo, pre-built binaries, or GitHub Releases (curl -sSL https://aof.dev/install.sh | bash)
  • Maintained under the agenticdevops organization on GitHub; documentation at docs.aof.sh

See Also

References