oritera/Cairn

What it solves

Cairn is a general-purpose problem-solving engine designed for tasks that involve a directed search through a near-infinite state space where the starting point and goal are known, but the path to reach the goal is unknown. While validated primarily on AI penetration testing, it is applicable to vulnerability research, mathematical proofs, and CTF challenges.

How it works

Cairn uses a Blackboard Architecture with a fact-intent graph to coordinate agents. It operates on three primitives:

• Facts: Confirmed objective findings.
• Intents: Declared directions for exploration.
• Hints: Human judgment injected into the system.

Agent Workers run an OODA loop (Observe, Orient, Decide, Act) to explore the state space. They coordinate through the shared board (stigmergy) rather than direct communication. The system consists of a Cairn Server (maintaining the graph), a Dispatcher (scheduling tasks and managing containers), and Worker Containers where multiple agents run concurrently.

Who it’s for

It is designed for security researchers, penetration testers, and anyone tackling complex, state-space search problems that require autonomous AI agents to discover a path to a specific goal.

Highlights

• Role-less Agents: Workers have no fixed roles; tasks are generated at runtime based on the current state of the graph.
• Bespoke Architecture: Uses a Blackboard Architecture to avoid information silos and direct agent communication.
• Proven Performance: The only team to achieve "AK" (solving all 54/54 problems) in the Tencent Cloud Hackathon AI Penetration Testing Challenge.
• Flexible Execution: Supports both Docker-based container execution and a local mode for running workers directly on the host.