The Village

The Village is the bird’s-eye view of everyone you employ. The top of the screen is a painted map: each Company is a neighborhood, each worker a little hut, with streets, a wall, a Library, and the Kafenio in the middle. The rest of the screen is the AI Tribe — the security and social layer that keeps every worker honest: the defences that guard them, the incident log of what’s been caught, and the Tribe Doctrine that’s woven into every worker’s prompt.

How to get here

Click 🏘 The Village (the Tribe / Village entry) in the sidebar.

The painted map

Switch between 🗺 Map and 📋 Roster with the toggle (your choice is remembered). The map is a top-down, hand-painted village you can drag to pan and pinch / ⌃-scroll to zoom:

• Neighborhoods are Companies; huts are workers. A hut with rising smoke was active recently; a lit candle marks a proactive (heartbeat) worker.
• Click a hut to open that worker’s profile, click a Company sign to enter the neighborhood, and click the Kafenio to step into the tavern or the Library to open it.
• Two standalone buildings sit at the corners: the Designer’s Corner (your visual artist) and the Chief Inspector’s Watch (the tribe auditor). Workers not bound to a Company gather in a wanderers’ camp below.
• The scene has a day/night mood that shifts with the real clock; lanterns glow at dusk.

A view, not extra setup

The map just visualises the workforce you already have — adding a Company or worker elsewhere makes a new neighborhood or hut appear here. If the painter ever hiccups, the screen falls back gracefully and the Roster view always works.

The AI Tribe: defences

Below the map, 🛡️ Defences are the “village wall” — concentric safeguards every worker runs behind: a system-prompt anchor users can’t override, tool approval gates on sensitive actions, audit-everything logging, an exfiltration detector, a prompt-injection classifier, and a budget circuit breaker. Click any card to drill into its incident feed; for the two heuristic detectors you can even paste a string and watch it fire.

Incident log

The 🚨 Incident log is the village bell tower: every prompt-injection, social-engineering, or exfiltration attempt the defences caught, freshest first, with a severity tag and a Resolve button. A quiet wall is a good wall.

The Tribe Doctrine

Open ⚙ Village settings to edit the Tribe Doctrine — your workspace “constitution.” It’s free-text (Markdown) that gets folded into every worker’s system prompt as authoritative house rules: things like “every worker is GDPR-aware; never log raw emails” or “reply in Greek to Greek-named users.” The same dialog sets the detector sensitivity (low / medium / high) and which event kinds escalate to an incident.

Write the rules once, everywhere obeys

The Doctrine is the highest-leverage setting in the app: one paragraph here shapes how all your workers behave, across every Company. Start small — a few clear, non-negotiable lines — and let it grow.

The canvas is also being expanded into a multi-instance world — a future map that stretches across your whole Fleet, not just this machine.

In practice

The Doctrine is the highest-leverage text in the app. It’s folded into every worker’s system prompt, across every Company, so a rule written once here doesn’t need repeating in a single persona. Concrete examples of what belongs in it:

• A non-negotiable data rule — “never log raw email addresses; refer to people by first name only.” Written here it binds workers you haven’t created yet.
• A house standard for output — “always cite the Library item you drew a claim from; if there isn’t one, say the claim is unsourced.” That single line changes how every report reads.
• A language or locale rule — “reply in Greek to Greek-named correspondents.”

Start with a few clear lines rather than a policy document; the Doctrine costs prompt space on every run, and vague rules produce vague compliance.

Read the incident log occasionally even when it’s quiet. It records prompt-injection, social-engineering and exfiltration attempts the defences caught — mostly relevant once workers are reading web pages, email or other untrusted content, where a hidden instruction can arrive inside data. An empty log after months of web-facing work is genuinely good news; you just want to have looked.

The detectors are testable. For the two heuristic ones you can paste a string and watch it fire, which is the only honest way to know what your chosen sensitivity actually catches before you rely on it. Worth five minutes if you’re raising or lowering the setting.

The map is a view, not a configuration surface. Nothing here creates or changes workers — neighborhoods and huts appear because you added Companies and workers elsewhere. Its practical use is spotting the shape of things: which Company is oversized, who’s sitting in the wanderers’ camp with no team, which huts show no smoke because that worker hasn’t run in weeks.

🎓 Learn it hands-on: Run many companies

Terms in this page

• The Village — the workforce-wide view: a painted map plus the Tribe’s security and social layer.
• Tribe — the collective of every worker across every Company, with shared defences and doctrine.
• Neighborhood — a Company, drawn as a cluster on the map.
• Hut — a single worker on the map.
• Wanderers’ camp — where workers not assigned to any Company are shown.
• Heartbeat / proactive worker — a worker set to act on its own schedule; shown with a lit candle.
• Kafenio — the workers’ tavern/watercooler, reachable from the map’s centre.
• Pan / zoom — drag to move the map, pinch or ⌃-scroll to scale it.
• Defences — the layered safeguards every worker runs behind (the “village wall”).
• System-prompt anchor — a non-negotiable safety preamble at the top of every worker’s prompt that user input can’t override.
• Tool approval gate — a checkpoint that pauses a sensitive action (shell, file write, browser typing) until you approve it.
• Audit log — the record of every state-changing action, kept so an agent can always be retraced.
• Prompt injection — a hidden instruction smuggled into content or tool results that tries to hijack a worker (“ignore previous instructions…”).
• Social engineering — manipulating a worker into bypassing its safety rules.
• Exfiltration — coaxing a worker into leaking secrets, credentials, or private data.
• Classifier / detector — a check that scans messages or content for known attack patterns and flags them.
• Budget circuit breaker — a hard spend ceiling per Company that stops runaway costs.
• Incident — a caught security event, recorded in the log with a severity and resolve state.
• Tribe Doctrine — the workspace-wide rules (Markdown) folded into every worker’s system prompt.
• System prompt — the instructions that define how a worker behaves; the Doctrine is appended to it.
• Detector sensitivity — how aggressively the passive defences flag unusual activity (low, medium, high).
• Fleet — your set of connected askTheodor machines; the map is being extended to span all of them.