Influences
The projects and ideas that shaped Guildhall’s design.
Jake Van Clief — Folder as Workspace
The insight that folders and markdown files replace agent frameworks. Three-layer routing: a top-level map the AI always reads, workspace-level context files loaded per task, and skills/tools plugged in at specific points. No databases, no Python frameworks, no agent orchestration libraries. Just English in text files.
Key contribution to Guildhall: the configuration layer is literally folders and markdown. Each Quorum seat is a folder with a SEAT.md. Each Lugh pipeline stage is a folder with a STAGE.md. The orchestrator reads these folders — it doesn’t interpret a DSL or parse a config database.
MemPalace — Navigable AI Memory
Wings, halls, rooms, tunnels, closets, drawers. A spatial metaphor for organizing AI memory so that retrieval is structurally guided rather than flat semantic search. The key innovation is that structure itself improves retrieval — wing + room filtering yields 34% better recall than searching everything.
Key contribution to Guildhall: the memory architecture. Each wing accumulates institutional knowledge. Each room (seat/stage) builds expertise over time. Tunnels connect shared concepts across workflows. Verbatim storage in drawers means nothing is lost.
Also: the specialist agent concept — agents that maintain their own diaries and build domain expertise across sessions — maps directly to Quorum seats accumulating pattern knowledge over repeated runs.
No Wrong Door (ACL/HHS) — Routing Philosophy
A U.S. government services design principle: regardless of who you are or which door you walk through, you get routed to the right service. 56 states and territories, 1,322 access points, all connected.
Key contribution to Guildhall: the routing layer. Users don’t need to know whether their request is a Quorum job, a Lugh job, or a memory search. They walk in any door and the system routes them. The internal architecture is invisible.
Paperclip — Multi-Agent Orchestration
Open-source orchestration for teams of AI agents. Org charts, budgets, governance, goal alignment, audit logging. Task checkout is atomic, preventing double-work. Tasks carry full goal ancestry so agents see the “why.”
Key contributions to Guildhall: the audit trail pattern (structured logging of each seat’s reasoning chain), atomic task management (seats don’t duplicate work in parallel batches), and goal ancestry (each seat receives the original prompt plus framing context, not just a narrow instruction).
What Guildhall doesn’t take from Paperclip: the always-on autonomous company model. Guildhall is deliberative and batch-oriented, not a 24/7 operation.
OpenClaw — Personal AI Agent
The “hands” layer — an agent that actually does things on your computer. Files, shell, browser, email, APIs. Skills-based architecture with persistent memory.
Key contribution to Guildhall: a potential execution layer. If Guildhall’s output is “do this thing,” OpenClaw could be the hands. Lugh’s TTS rendering, for instance, could delegate to an OpenClaw skill. Not a core dependency — a convenience integration for when Guildhall needs to act on its conclusions rather than just report them.
De Bono’s Six Thinking Hats
The original framework that Quorum expands. Six cognitive roles applied to a problem: facts, emotions, caution, benefits, creativity, process. Quorum’s contribution is recognizing that the Black Hat (caution) is underspecified — it collapses three distinct failure modes (rational exploitation, asymmetric extraction, active adversarial opposition) into one bucket.
The Tenth Man Rule (World War Z)
Structurally mandated dissent. If everyone in the room agrees, someone must argue the opposite. Not because contrarianism is valuable in itself, but because unanimous agreement is a signal that the group may have missed something. In Quorum, the Tenth Man argues against whatever consensus the other 11 seats reached.