AI-OPS Documentation

What is AI-OPS

AI-OPS is a completely 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 can also be used directly as a Python library (see Run the Agent Programmatically) for use-cases such as benchmarking or automation.

What AI-OPS isn't

A banana.

Start here

New to AI-OPS? Getting Started takes you from nothing to a running API with the CLI connected, and your first task sent.

Deploy and operate

For running AI-OPS and using it day to day.

• Getting Started: deploy the API from source and connect the CLI.
• Run the API Server: configuration, exposing beyond localhost, driving the API without the CLI.
• Use the CLI: flags, config file, in-app keys, slash commands, modes, resume.
• Model Selection: choose a model that works, including self-hosted and OpenAI-compatible endpoints.
• Configuration: every API, CLI, and agent setting.
• Command Policies: allow-list what the terminal tool may execute.

Extend

For adding capabilities or embedding the agent in your own code.

• Add a Tool: give the agent a new action.
• Add a Skill: give the agent task-specific instructions on demand.
• Add a Command Policy: write a custom terminal admission rule.
• Run the Agent Programmatically: use ai_ops.core directly, including the synchronous execution model for benchmarks and prompt optimization.

Reference

• API Reference: endpoints, events, schemas, configuration, auth.
• Built-in Tools: the tools the agent can be given.
• Context Management: how the context window is compacted.
• Storage Layout: the base directory, sessions, and the session store.
• Bundled Skills: the skills that ship with AI-OPS.