Attesté is a SaaS company operated day-to-day by a single person and an AI agent on a custom-fit Substrate. This is a guided flight through its actual cockpit — every screenshot below is the production system, not a mock.
The thesis: one person can run a company — if every consequential action passes a gate a human can trust.
The Command Centre is a local desktop application. It runs on the operator's own machine, was never deployed to the public internet, and keeps its credentials in an encrypted local store. The substrate stays home. What follows tours its four wings: Studio, Growth, Intelligence, System.
The Video Studio accepts a persona, a hook format and a content pillar — there are no free-form briefs, because free-form briefs are how a brand voice dies. The blockers panel refuses to generate until authentication, trained identity models and approved voices exist. Fail closed, not open.
A Soul-anchored video body. Voiceover and brand-wrap assembly. A vision model interrogating the stills — would a collector screenshot this? Then the anti-slop judge. Whatever survives still lands in the Content Queue for a human to approve, and only then schedules out to the calendar.
Every run shows its estimate and its expected keeper rate — roughly one keeper in four, budgeted at two to three generations. An operator who can see the cost of taste can afford to have it.
Auto-publish stays disarmed until the anti-slop judge agrees with a human reviewer on more than 90% of ~30 labelled posts. The calibration screen shows zero posts labelled and no agreement score — because none has been earned yet, and the system refuses to display a placeholder number.
Provenance irony — never fake something real on an authenticity brand; a staged "real-world" moment fails outright because the medium would contradict the message. Automation meta — never make the AI the story; copy that winks at its own automation breaks the fourth wall the brand depends on.
The personas are trained identity models — a Soul binds a likeness, so training one cannot be triggered from the cockpit at all; the train function throws by design. The only path runs through the provider's per-character web consent flow, completed by a person. The interface documents its own restraint.
From not started through research, first contact, follow-ups, sample kits and meetings to signed. An enrichment scheduler works the backlog on a daily budget, scoring and tiering every gallery; the strategist's Playbook turns scores into each morning's calls.
The principle that governs the studio governs growth: drafted emails land in an awaiting-review queue, and a person presses send. AI speed where speed is cheap; a human signature where a relationship is at stake.
"We do not display a placeholder number." — the calibration screen, refusing to fake its own evidence
The cockpit is the visible tenth. Beneath it, the Substrate holds what the business actually knows — structured, queryable, on hardware the operator controls.
Every artwork, artist, gallery and persona — relationships, not rows.
Every prospect and conversation, and where each sits in the buying journey.
MRR, subscriptions, churn, lifetime value, runway — reconciled.
Every campaign and result: what worked, what didn't, what to repeat.
POPIA inventory, processor agreements, terms, trademarks.
Shipped, in-progress, deferred, killed — with the rationale kept.
Each morning the loop runs again — slightly smarter than yesterday, because yesterday was captured.
Overnight signals become a briefing: what changed, what needs a human, what's drifting.
Approve, modify or kill the proposed actions. Minutes, not meetings.
Everything that doesn't need a signature runs; drafts queue in the outbox.
Decisions, conversations and changes land back in the substrate.
Every command captured — append-only, hash-chained, on hardware the operator controls.
The substrate refuses to type passwords or 2FA codes. Keys stay with the human.
Destructive or outward actions queue for explicit approval. Nothing auto-fires.
The operator-substrate pattern fits any business whose day-to-day is information-handling: services, advisory, software, content, finance. The diagnostic maps it to yours — plain questions, fixed quote at the end.