AI-OPS

LLM-assisted penetration-testing agent. Bring your own LLM, run the API, drive it from the CLI.

Table of Contents

• Overview
• Features
• Architecture
• Quickstart
  • Requirements
  • Run the API
  • Connect the CLI
• Supported Models
• Documentation
• Use as a Library
• Contributing
• Disclaimer
• License

💡 Note: AI-OPS is an active research prototype.

Overview

AI-OPS is an open agent for penetration-testing and cybersecurity tasks. You run the agent behind an HTTP API, in a container with the offensive tooling preinstalled, and drive it from a terminal client.

It is built to be model-agnostic: it uses litellm under the hood, so you can point it at a hosted provider, an OpenAI-compatible endpoint, or a self-hosted model. The design targets medium-sized LLMs, so it stays usable without frontier-scale hardware.

This is mostly an experiment tool, don't expect it to replace any actual competence in pentesting/cybersecurity.

Features

• Bring your own LLM. Any provider litellm supports, including OpenAI-compatible and self-hosted endpoints.
• Skills. Reusable, bundled playbooks the agent can load for common tasks. Can extend with your own.
• Guarded command execution. A terminal tool runs commands under configurable allow-list policies, with confirmation for anything not on the list.
• Context management. Layered context views keep long sessions within a model's window (under analysis).
• Observability. Optional tracing through MLflow.
• Two ways to use it. Drive it interactively from the CLI, or call the API (and the Python library) directly for automation and benchmarking.

Architecture

AI-OPS has two parts:

• API: a containerized HTTP server that owns the agent, the tools, and conversation state. This is what you deploy.
• CLI: a terminal client that connects to the API and gives you an interactive session.

The LLM is external and yours to choose.

Quickstart

Requirements

• Docker to run the API.
• The CLI, either the prebuilt binary or Node 22+ to run it from source.

Run the API

Pull and run the image, passing your model configuration as environment variables:

docker run --rm -p 8000:8000 -e AI_OPS_MODEL=openai/your-model -e LLM_API_BASE=http://your-endpoint/v1 -e LLM_API_KEY=your-key -v ai_ops_data:/home/aiops/.local/share/ai_ops ghcr.io/antoninolorenzo/ai-ops:latest

• AI_OPS_MODEL is the only required setting (a fully-qualified litellm model id). LLM_API_BASE and LLM_API_KEY are optional and depend on your provider.
• The volume persists conversation history across runs.
• Bound to 127.0.0.1 the API runs without auth for local use. If you expose it on a network, set AI_OPS_AUTH_TOKEN. See Run the API Server for more details.

Connect the CLI

Download the latest ai-ops-cli from Releases, then:

chmod +x ai-ops-cli
./ai-ops-cli --base-url http://127.0.0.1:8000

Flags, the config file, slash commands, and resume are covered in Use the CLI.

Documentation

Full documentation lives at antoninolorenzo.github.io/AI-OPS:

• Getting Started
• Run the API Server
• Use the CLI
• Configuration
• Extending AI-OPS (tools, skills, policies)

Use as a Library

The agent can be driven directly from Python for automation or benchmarking, without the API. See Run the Agent Programmatically.

Contributing

Contributions are welcome. See CONTRIBUTE.md for setup, testing, and documentation guidelines.

Disclaimer

AI-OPS is designed as an ethical hacking tool intended for academic and educational purposes only.

The creators and contributors of AI-OPS are not responsible for any misuse of this tool. By using AI-OPS, you agree to take full responsibility for your actions and to use the tool in a manner that is ethical, legal, and in accordance with the intended purpose. The creators are not liable for any damages or legal repercussions resulting from the use of this tool.

License

Released under the MIT License.