Guide

Guildhall is a local AI agent harness for software projects. Start with the first-read pages if the words "agent harness," "blueprint," or "guild" are not already obvious.

First Read

• What Guildhall is — the basic concept, including why the product uses the name Guildhall.
• Start here — install, open one project, and run a small task.
• How Guildhall works — how planning, context, workers, reviewers, gates, and memory fit together.
• Core concepts — the vocabulary in one place when a term gets fuzzy.

After that, the docs follow the product domains you actually touch:

• Projects are the repos Guildhall can see.
• Tasks are the pieces of work you ask it to move.
• Specs and levers shape how work is planned, reviewed, recovered, and learned from.
• Blueprints and inspections are how Guildhall keeps work coherent without turning every run into a hidden chat transcript.

Setting up your first project? Start with Start here.

Projects

• Projects and work — service home, project cards, and the project shell.
• Project files and workspace state — what lives on disk.
• Running Guildhall — browser controls first, CLI commands when you need them.
• Guildhall app reference — screen-by-screen details when you need a specific app page.

How Guildhall Works

• How Guildhall works — the system model: survey, blueprint, context, workers, reviewers, gates, and memory.
• How Guildhall builds — the construction model behind planning, implementation, review, and release.
• Research-backed design — the research and practice ideas behind smaller tasks, bounded questions, visible uncertainty, and audit trails.
• Agent context — what agents receive before they act.
• Corpus Map — how Guildhall indexes a project without dumping the whole repo into every prompt.
• Memory, learning, and recovery — how Guildhall learns reusable habits without turning them into mystery behavior.
• External agents and MCP — how Codex, Claude Code, and other MCP-aware tools can read Guildhall project context.

Tasks

• Pressure-Test Intake — how broad requests become buildable specs without turning uncertainty into invisible guesswork.
• Task lifecycle — how a task moves from idea to done.
• Cleaner project notes — how Guildhall keeps shared project notes readable without losing the receipts.
• Git Story Closure — how Guildhall shows whether completed work is dirty, local, pushed, in a PR, merged, local-only, or deferred.

Specs And Levers

• Onboarding and levers — how behavior settings get proposed and approved.
• How Guildhall routes work — how Guildhall routes work without making you manage a steward roster.
• Agents and models — roles and provider assignments.
• Open model recommendations — tested open-model lanes and how to compare candidate models.