How it works

UnitWell works the portfolio on its own, and stops at the line where a decision costs someone money. This is that loop — what runs unattended, what waits for you, and what you see when it does.

1. What the agent does unattended

A sweep runs over the whole portfolio and reads every live lease, its ledger, and its open work. Then it does the part that does not need a judgement call:

Posts each period's rent to the lease ledger — once, on a duplicate-safe write, whether the lease bills monthly, semi-monthly, or biweekly.
Answers residents about their own balance, lease dates, and open repairs, reading their record rather than guessing. A resident can only ever see their own unit.
Takes in a new maintenance request, classifies it, and matches it to a vendor from your list — manager-starred vendors first.
Sends the rent reminders on the collection ladder, inside your quiet hours.
Drafts the things a person will edit: renewal terms, listing copy, the owner's summary.
Raises a flag when something is missing rather than inventing a value — no late-fee policy configured, no insurance on file, a work order that has sat scheduled too long.

None of that moves money out of an account, changes what someone owes under their lease, or penalises anyone.

2. What lands in the approval queue

The moment an action would move money, change a lease, or penalise a resident, the agent stops and files a task instead. The task carries the amount and the reasoning; it does nothing until a person acts on it.

Late fees and returned-payment fees.
Payment plans.
Lease renewals, move-outs, and any change to a lease term.
Security-deposit returns.
Applicant approvals and denials.
Vendor dispatch above $400. Below that a quote is dispatched and booked to the property's expenses automatically.
Recurring ancillary fees above $400 — parking, storage, pet, amenity. Below that they post to the ledger automatically.
Posting a utility rebill across a property's leases.
Changes to your org's own policy settings.

The $400 line is one number in one place. Above it, a human signs off; below it, small routine repair work is not held up waiting for you.

3. Rent follow-up, concretely

Chasing rent is the job most owners want off their plate and the one most likely to go wrong if it is automated carelessly. So it is not a loop that gets louder — it is a fixed ladder anchored to the period's due date, and it is the same ladder whether it is reached by the daily sweep or by a payment landing.

Three days before rent is due
A short reminder that rent is coming.
On the due date
If a balance stands, a due notice.
When the grace period runs out
A follow-up. Grace is the longer of your state's statutory grace and whatever you configured — never the shorter.
Three days after grace
A firmer notice.
Five days after grace
A late fee is PROPOSED — filed as an approval task, capped by statute and by the lease, and applied only if you approve it.

The brakes on it

One rung per run, and never a skipped rung. If the system were down for a month, the resident still gets the reminder, then the follow-up, then the firm notice before a fee is ever proposed. A catch-up must not become an ambush.
Never two automated resident messages on the same lease within twenty hours.
Nothing goes to a resident outside your organisation's quiet hours — including anything marked urgent. A message about money at 3am is not an emergency to the person receiving it.
An active or pending payment plan pauses the whole ladder. The agent does not escalate on someone it just offered help to.
A late fee that has already been proposed once is never proposed again for that lease — including one you declined. You said no; it does not ask again.
The balance being settled clears the ladder outright.
A lease that runs late three times in six periods is flagged to you as a pattern, not escalated harder.

4. What a manager sees

An approval queue, with the reasoning attached

Every proposal arrives as a task carrying what it is, who it affects, the amount, and why the agent raised it. You approve or decline from there.

Approving is the step that applies it

Nothing is pre-applied and then rolled back. The effect runs on approval. If a precondition blocks it, the task goes back to the queue marked as not completed rather than being reported as done.

A trail with names on it

Approvals and declines are both written to the audit trail with who did it and when. Declining a maintenance dispatch also marks the work order itself declined, so the decision is visible where the work lives.

A digest instead of a dashboard vigil

After each sweep the manager gets a summary of what ran, what was proposed, and what is still waiting. Owners get their own plain-language version.

What this is not

The language model brain is off by default. Unless it has been switched on for your deployment, the agent answers from deterministic rules rather than a model — and either way the approval line above is the same.

Several outside integrations are still simulated in the current build, including card and bank payment rails, e-signature, SMS, and applicant screening. We list them, one by one, on the disclosures page.

Nothing the agent produces is legal, tax, or financial advice, and it is built to decline requests to draft notices or quote statutory limits and hand those back to a person.

Watch the loop run

The tour walks a proposal from the sweep to the approval queue. About a minute, no sign-up.

See the live demo Read the FAQ