All updates
Feature
Assistant: it learns from your thumbs, and access codes are shown in full
A thumbs down can say in one tap what was wrong. When you specify a property after an ambiguous question, the assistant remembers it for your whole company. And access instructions no longer show “[masked]”: codes are displayed, without ever being sent to the AI.
The assistant keeps nothing by itself from one conversation to the next. What learns is the software around it, from your feedback.
- Thumbs down: six chips appear (Wrong property, Wrong person, Misunderstood, Incomplete, Incorrect, Other). One tap is enough, the free comment remains possible, and everything is optional.
- When the assistant asks “which one?” and you choose, the word you had written (for example “Oberkampf”) now means that property for your whole company. A banner says so the first time; afterwards it works silently.
- When a successful rephrasing follows a thumbs down, the pair is kept and reused when a question looks similar. An example that then receives two thumbs down is set aside on its own.
- Settings › Assistant: the card “What the assistant has learned from your corrections” shows these links; “Forget” removes them, after confirmation.
- Access codes: a property's instructions used to show “Door code: [masked]”. Codes, phone numbers and addresses are now shown in full to you. The AI, for its part, still only receives a placeholder in their place: the value never leaves the software.
Every Monday, after the automatic tests, a review of the week's thumbs down suggests rules to our team to add to the assistant's instructions. Nothing is applied without a human click, and if a rule lowers the test score, the previous version comes back on its own.