No roles, no RAG, no plan: just a blackboard and a goal

What it does

Cairn is a problem-solving engine that looks for a path from a known starting point to a defined goal through an otherwise unknown state space. It keeps a shared graph of facts, intents, and human hints on a blackboard, dispatching agent workers to explore until the goal is met. So far it has only been validated on penetration testing and CTF challenges.

The interesting bit

The agents never talk to each other directly. They coordinate through stigmergy—reading and writing to the same blackboard—so tasks emerge from the graph state instead of fixed workflows. According to the project’s own account, this architecture cleared all 54 challenges in a Tencent Cloud hackathon on its first live run, brought online at 4 AM on race day with no prior tuning, RAG, or MCP tooling.

Key highlights

• Uses only three blackboard primitives: Fact, Intent, and Hint
• Agent workers run an OODA loop with no fixed roles or inter-agent messaging
• Supports worker backends including Claude Code, Codex, and Pi
• Took 3rd place among 610 teams in a Tencent hackathon while being the only team to solve all 54 targets
• Dual-licensed under GNU AGPLv3 for personal use, with a commercial license available

Caveats

• General-purpose claims are unproven: the README explicitly states penetration testing is the only validated domain so far.
• Platform requirements: it currently requires macOS or Linux, Python 3.12 or higher, and Docker.

Verdict

Worth exploring if you are building agent systems and suspect that rigid roles and workflows are getting in the way. Skip it if you need a proven general-purpose solver or Windows support.