AI statement
Agents propose. A person decides.
How AI works in Sentinel, and what we have not measured. As of 5 November 2026.
Who decides
Every AI proposal arrives as a Review Card with what will happen, why, and the rule. Two levels exist today: Suggest, where it drafts and you do it, and Ask, where it does it after you approve. There is no setting that lets AI act by itself.
- Sentinel moves no money.
- Applicant decisions are a person's, with a name and a time on the record.
- Rent suggestions stay suggestions. They say it is your decision, and have no path to your PMS.
- Review replies are drafted. A person posts them.
AI does not talk to your residents or applicants
Sentinel has no bot that speaks to residents, applicants or prospects. If you run a leasing assistant, it stays yours. Agents are named by job, such as AP agent, never with a human name.
Where it can help
Reading contracts and statements, suggesting GL coding, drafting review replies and messages, ticket briefings and knowledge articles, comp pages that cannot be read plainly, and the Rent Advisor. People has no AI, by design. Confidential support cases never go to AI.
Off works
AI needs a key on the server, and a company can switch it off. With AI off, every page still works, and says so where it matters. The sample portfolio never calls AI.
What it costs
Each company has a daily spend cap, set during setup and changeable later. When it is reached, AI pauses until the next day and an item goes on Today. Admin shows each run's purpose and cost.
Your data
Calls go through one gateway to Anthropic's API. Text is redacted by default, including contact details on contract and statement reading. Prompts are not stored by us. We do not use your data to train anything. Your data does not feed another company's answers.
Fair housing
A fair-housing rule goes into AI prompts, and returned wording goes through the same check as typed wording. Those checks catch common problems, not all. A person reads everything that goes out. Memories on a renter's record refuse wording about who someone is.
What we have not measured
We test the checks around the AI, on local stand-ins. As of this date we have not measured how well it reads real contracts or real statements, so please treat each result as a draft. Confirm screens show the quote and how sure the reading was, so you can check. When we have numbers, they will carry a date and a sample size.