oritera/Cairn

🧭 Cairn: General AI State-Space Search

Solves problems with unknown paths. First validated on autonomous penetration testing.

Cairn is a general-purpose AI state-space search engine. Defines no roles, no workflows. Given an origin and a goal, searches for a path through an unknown state space. Penetration testing is the first domain validated. Built on a Blackboard Architecture with an explicit fact-intent graph. Three primitives: Fact (confirmed finding), Intent (direction of exploration), Hint (human judgment). Agents run an OODA loop — observe the full graph, orient to current state, decide next intents, act to explore. Workers have no fixed roles. Tasks generated at runtime from the graph's current state. Agents coordinate exclusively through the shared board (Stigmergy). No direct communication. No information silos.

✨ Highlights

• Blackboard Architecture: Shared graph with three primitives — Fact, Intent, Hint
• OODA Loop: Agents observe the full graph, orient to current state, decide next intents, act to explore
• Stigmergy Coordination: Agents communicate only through the shared board. No direct communication. No information silos.
• Runtime Task Generation: Tasks generated from the graph's current state, not from predefined job descriptions
• Multi-Backend Support: Claude Code, Codex, and Pi worker backends
• Hackathon Record: Solved all 54 problems at Tencent Cloud Hackathon AI Penetration Testing Challenge. Placed 3rd.

💡 Why it matters

Cairn abstracts any problem with a clear start, clear goal, and unknown path as state-space search. This shape applies beyond penetration testing — vulnerability research, CTF challenges, mathematical proof. Blackboard architecture eliminates role rigidness. Agents coordinate through shared facts, not handoffs. The hackathon run required zero training, zero tuning, zero domain-specific tooling. No MCP tools, no RAG, no predefined agent roles.

#AI #StateSpaceSearch #PenetrationTesting #BlackboardArchitecture #TiniX