Office Town

AI agents that work like a team. Four buildings, four roles, a wiki-backed substrate. Your agents stop guessing — they know the work.

Office Town is a markdown methodology for AI agent fleets running on Goose (or any Open Plugin Spec host) with a Cloudflare Workers backend.

your-town/
├── buildings/
│   ├── office/        # @boss — routes work, holds the thread
│   ├── library/       # @librarian — extracts + curates the wiki
│   ├── workshop/      # @worker — deep work
│   └── lookout/       # @scout — outward scanning
└── wiki/              # the shared substrate (11 default collections)

---

Get started

You already have a capable AI agent — Claude Code, Goose, OpenAI Codex, Aider, Cline, etc. Don't read setup docs. Tell your agent to install it.

👉 Open INSTALL.md — three prompts to paste into your agent:

1. Full install — deploys Cloudflare backend + sets up plugin + first town
2. Cloud-only — just deploy the backend, you wire up the rest
3. Connect-existing — wire a new machine to an existing town

Or visit officetown.au for the same prompt in a copy-button-shaped form.

If you don't have a capable agent (or want a deterministic manual path), see SETUP.md.

Or download Office Town Desktop

A pre-configured macOS .app — signed + notarised, no warnings:

https://download.officetown.au/mac

Then run the same agent-install prompt from inside it. The .app is rebranded Goose with our icon + bundle ID, so any agent on macOS understands what it's looking at.

What you get after install

• 4 addressable role agents: @boss, @librarian, @worker, @scout
• A wiki MCP backed by FTS5 + Vectorize hybrid search
• Files MCP, publish MCP, cron routines, dashboard
• A pluggable role-pack catalogue: pack-startup, pack-design, pack-hosting, pack-wordpress, pack-business, pack-cloudflare, pack-comms, pack-knowledge
• Cost: ~$2-5/month on Cloudflare at typical SMB volume

How it actually works

Primitive	Job
Town	Your installation — a folder that defines who, what, where
Place (Building)	A workplace — Office, Library, Workshop, Lookout
Role	A specialist agent — addressable by name
Task	A discrete unit of work — routed from boss to a specialist

Each role is a markdown file with an AGENTS.md that Goose loads on session start. The wiki is the shared memory — entries have a universal sextet of frontmatter fields (slug, kind, created, last_updated, last_edited_by, last_change_summary) plus collection-specific fields.

Deep dive: METHODOLOGY.md.

Repo layout

office-town/
├── README.md           ← you are here
├── INSTALL.md          ← agent-paste install prompts (primary path)
├── SETUP.md            ← manual fallback (deterministic, no agent needed)
├── METHODOLOGY.md      ← the why and how
├── AGENTS.md           ← town-level standing orders (Goose auto-loads)
├── buildings/          ← 4 building folders, each with its own AGENTS.md
├── roles/              ← core role definitions
└── LICENSE             ← MIT

Related repos

Repo	What
office-town-cloud	Cloudflare Workers backend (wiki/files/publish/cron MCPs)
office-town-plugin	Open Plugin Spec plugin (roles + skills + recipes + hooks)
office-town-desktop	Custom Distribution of Goose Desktop (signed Mac .app)
office-town-pack-*	Role packs: startup, design, hosting, wordpress, business, cloudflare, comms, knowledge

Licence

MIT. Use, modify, redistribute. © 2026 Jezweb Pty Ltd.