oritera/Cairn

What sits underneath most agentic AI systems is a planning layer that needs to navigate from a starting condition to a goal through a sequence of decisions. This library formalizes that problem as state-space search and hands you a configurable engine to run it. The authors stress-tested the abstraction on autonomous penetration testing, which is one of the harder planning domains you can choose, and the core idea held up well enough to share publicly. The appeal for builders is the generality. If you can model your problem as a graph of states with defined transitions, Cairn can, in theory, drive the search. That covers a surprising range of agentic workflows: multi-step automation, game-playing agents, workflow orchestration, or anything where a brute-force LLM prompt chain feels sloppy. The honest reservation is that this is early-stage research code, not a polished SDK. Expect to read the source, patch rough edges, and contribute back if you want it production-ready. -> Best for: AI engineers prototyping agentic planning systems who want a principled search backbone rather than ad-hoc prompt chaining.