A command centre for the one-person company

Run your empire with agents you can trust.

Empire OS turns the businesses you run into divisions, each with its own AI agents that do real work in your real files — researching, writing, shipping — while you watch every move on a native macOS dashboard. Every change is checkpointed. Anything can be undone.

Native macOS · runs on your machine · bring your own AI keys — pick the models, control the bill

AI agents are powerful. Unsupervised, they're a liability.

Chat windows forget. Scripts run blind. Most "agent platforms" ask you to trust a black box with your actual files and give you nothing when it goes wrong. Empire OS was built the other way around: observability and reversibility first, autonomy second.

How it works

Five ideas, one operating loop.

DV

Divisions

Your businesses and projects — each with status, health, revenue targets, its own colours, and a folder of documents that is its knowledge base.

TK

Tasks

The work queue, classified agent-ready or needs you. Due dates flow into Apple Calendar and Reminders — and checking off a reminder closes the task.

AG

Agents

Standing missions with scoped tools and schedules. Run on demand or hourly-to-weekly, even while the app is closed.

CP

Capabilities

SOPs and skills discovered from your folders, MCP servers, and API credentials injected at run time — never shown to the model.

BR

The Brain

A local engine on your Mac that runs agents, fires schedules, writes your daily briefing, and streams every event to the dashboard live.

The daily loop

  1. Read the briefing. Every morning the Brain reads your whole portfolio and pins what needs attention, what moved, and what's next.
  2. Check the menu bar. Attention, up-next, and recent changes — one glance, no window.
  3. Work the queue. Do it yourself, or hand it to the agent and watch it execute live in Mission Control.
  4. Verify. Open any run's Changes panel: every file touched, line counts, one-click revert.

Why you can let it run

Built so the worst case is a git revert.

Jailed execution

File and shell tools resolve every path against your content root and refuse anything outside it — backed by an OS-level sandbox profile.

Everything checkpointed

Every run and doc rewrite is bracketed by git commits. The Changes panel shows exactly what was touched; revert restores the pre-run state.

Every tool call audited

Agents log each call to the live feed with its input and result. The trace reads like a flight recorder, not a mystery.

Per-agent permissions

Limit any agent to exactly the tools it needs — read-only research agents stay read-only.

Drafts, never sends

Mail integration composes drafts for your review. By design, the system cannot send a message on its own.

Bounded runs

Turn limits, command timeouts, and output caps mean a runaway agent stops itself.

The surfaces

Everything has a place.

Dashboard

The portfolio at a glance — a territory card per division with health arcs and agent pips, the pinned morning briefing, and a live feed where errors float to the top and agent activity collapses into an expandable trace.

portfolio

Mission Control

The live cockpit. Type a command, scope it to a division, and watch the agent think and act — every tool call streaming in. Replay any past run with its cost, duration, file changes, and revert button.

execution

Agent Studio

Describe a goal and the Studio drafts the mission and system prompt, scores capability, flags gaps, and suggests what to equip. Tool policy, schedule, and deploy — one surface. It will even audit your division's documents and propose rewrites you review as line diffs, with batch apply and undo.

design

Conversations

A persistent, tool-equipped chat per division. Sessions save automatically, resume with full context, and export to Notes or markdown.

thinking

Menu bar

Always present: attention items, what's next, recent changes. The empire runs even when the window is closed.

ambient
0
turn cap per run — agents stop themselves
0
of tool calls audited to the live feed
0
click to revert any run's file changes
$0
markup on AI — you pay your providers directly

Your models, your bill

No seats. No token markup. You hold the keys.

Empire OS doesn't resell AI — you plug in your own Anthropic and Gemini keys and pay your providers directly, at their prices. Pick the model for each job right in Settings: a frontier Claude model where agents touch your files, a fast Gemini model for briefings and document audits. The dashboard shows what every division spent this month, so the bill is never a surprise — it's a metric you manage.

Local-first

Your Mac. Your files. Your business.

The Brain runs on your machine — no third-party server holding your company. API keys live in a private file on your Mac and are never echoed back, never synced, never stored in a shared database. Model calls go to the providers you chose; everything else stays home. Multi-Mac setups share one database; your documents sync however you already sync them.

FAQ

Fair questions.

What do I need to run it?

A Mac on Apple Silicon (macOS 15+) and an Anthropic API key — paste it during the guided setup and you're running in about fifteen minutes. Add a Gemini key any time to route the lighter work to a cheaper model.

Why two AI providers?

Different jobs deserve different price points. Agent execution — the part that touches your files — runs on Claude, the strongest tool-using model family. Briefings, document audits, and command dispatch can run on Gemini Flash for a fraction of the cost. You pick both dials in Settings.

Can the agents touch anything outside my chosen folder?

No. Paths are resolved and refused outside the content root, and shell commands additionally run under a macOS sandbox profile that denies writes elsewhere. With no root configured, execution is refused entirely.

What happens when an agent gets something wrong?

Open the run, read the Changes panel, click Revert. Every run is bracketed by git checkpoints in a repo the agents can't touch. Document rewrites from the Studio have their own one-click undo.

Is my data sent anywhere?

Model calls go to Anthropic (that's the AI). Everything else — your database, your files, your credentials — stays on machines you control.

What does it cost to operate?

You pay your AI providers directly at their list prices — Empire OS adds no markup and meters nothing. Costs scale with how much you delegate, and the dashboard attributes every dollar to the division that spent it. Choosing smaller models for routine work keeps a busy portfolio surprisingly cheap.

Can it run on a Claude subscription instead of API keys?

Yes — a personal edition drives everything through a Claude Max login on your own Mac with no per-token cost at all. It's how Empire OS runs its own maker's company every day.

Not a demo

The empire it runs.

Empire OS isn't a mockup with sample data. It runs Ayres Capital Enterprises — a real portfolio of Canadian businesses — every day. These are the divisions on the dashboard. Each one is a live company; go have a look.

Player Piano AI

AI automation consulting for small business — and the maker of Maestro.

consulting

Millennial Press

An independent imprint publishing near-future fiction under the pen name Dewar Yearns.

publishing

ACE Home

Home services done right — landscape & interior design, maintenance, smart-home & security.

services

True North Supply

Curated goods from Canadian makers.

commerce

Skill Spark

Learn real-world skills, one hands-on micro-lesson at a time. iOS.

app

ACE Auto

Sourcing the collector cars of the 80s, 90s & 00s for the people who love them.

sourcing

Build your agentic organization.

Empire OS runs a real multi-brand portfolio every day — briefings at dawn, agents on schedule, every change revertible. Early access is opening to a small group of operators. Bring your keys; keep your margins.