Santara is certified AI-Native Operations Management Software.

The system does its work on its own schedule, not on the operator's prompt.

The Morning Brief is generated hourly per tenant on a scheduled state machine in the customer's own timezone, and reaches the operator by email and push without a session. Runs are recorded in engine_runs with their trigger payload, response and heartbeat, so a fortnight of unattended runs can be read back rather than asserted.

The system's primary output is a decision, not a dashboard.

The default surface is the brief, not the dashboard: one sheet naming what needs attention, what it is worth, and the cleaning that has to happen, each line carrying an action label and a destination (daily_report_items.action_label / action_url) and ordered by severity and sort order rather than by recency.

The system changes the world, not only the screen.

Smart pricing writes nightly rates through Channex to Airbnb and Booking.com; availability writes are clamped to the channel's own room count before they leave; turnover tasks and assignments are written to the operations board. Each night written is recorded in smart_pricing_days with its outcome — applied, would_apply, needs_approval, skipped_manual or unchanged — and the previous rate it replaced.

Every figure the system states traces to a record it can name.

Generated items carry lineage, not prose: ai_insights.source_data holds the rows an insight was derived from, and every brief line links to the property, reservation or task it concerns. Prices are reconstructable rather than reported — smart_pricing_days stores the whole multiplier stack for a night (base, season, day-of-week, event, pace, gap, last-minute) alongside the floor and ceiling that clamped it.

It says why, where the decision is made, and it can be overruled there.

The reason travels with the decision: the brief shows the multiplier chain behind a night's price where the operator sees the price, and each insight carries its recommended action and confidence. Overrides are honoured structurally — a rate whose price_source is manual is skipped by the engine and counted as skipped_manual, and a dismissed guest does not return to the attention queue until a new signal arrives.

Every autonomous action has a published limit, an audit record and a named owner.

The enforcement is real and the audit is real: pricing_guardrails holds per-property floors, ceilings, maximum daily and weekly movement, an auto-apply switch and an approval threshold; incomplete cost inputs raise costs_incomplete and fall back to a ratio floor rather than pricing against a cost of zero; withheld runs are recorded with the reason. What is missing is the third leg — there is no public statement a customer can read before switching the system on that says what Santara may do unattended and within which bounds.

With the model switched off, the system still produces a correct, safe result.

The engine that moves money is deterministic by construction: smart pricing is a multiplicative rule stack over data rows, with no model in the path, and upsell detection defaults to calendar-driven rules rather than inference. With inference unavailable the brief still assembles from operational state — arrivals, departures, cleaning, revenue, gaps — and loses only the written insight layer.

What the operator does with the output changes the output.

Operator response is captured and partly acted on: insight status moves through acknowledged, resolved and dismissed; a dismissed guest is suppressed from the attention queue until a newer priority message arrives; a manually priced night is left alone on every later run. What does not yet exist is a loop that learns — nothing in acceptance or dismissal rates changes ranking, thresholds or which detectors fire.

Remove the AI and the product stops being the product.

Inference sits on the promise, not beside it. Santara is sold as the system that hands an operator their day; remove the generated brief, the insight layer and the assistant and what remains is a channel-connected calendar — a category the company explicitly does not compete in. The AI is not a tier: it is on every plan, and there is no version of the product sold without it.