The assistant¶
The assistant answers questions about the BeCoMe method, about your projects and about how to use the app. It is an AI language model, not a person, and its answers can be wrong. Treat what it says as a starting point, and check anything that matters on the result page and in this documentation.
Where to find it¶
Sign in and click the assistant button in the top bar. It is a sparkle icon, and Assistant appears when you hover over it.
On a project's page, both that button and Explain this result on the result page open the panel for the project, and the panel header shows the project's name. Anywhere else in the app the panel is labelled Method and app. It has no project to look at there, so it answers questions about the method and about using BeCoMe. Moving to another page with the panel open does not change its scope. Pressing the assistant button or Explain this result again does, whether the panel is open or closed.
What it can answer¶
The method. How fuzzy opinions work, how the best compromise is computed, what the maximum error says. These are the questions the pages on this site answer too, so the assistant is mostly a faster way to the right paragraph.
Your project. Its opinions and its result, for example why the confidence is Moderate. The assistant sees only what you see as a member of that project. Every question you send from a project's panel carries that project's data, whatever the question is about. Your data says what that includes.
The app. How to create a project or invite experts. This works without a project selected.
The assistant only reads. It cannot change a project, an opinion or your account.
Limits¶
You can send up to 60 messages an hour. When you reach the limit, the assistant says how many minutes to wait.
An answer can take up to a minute.
In a long conversation, older messages are no longer sent with new questions. At most the last 20 messages go along, and fewer when the answers are long. The panel shows a line where the 20 end. The assistant stops taking older messages into account, so repeat anything it needs to know.
When something goes wrong¶
Message limit reached. You have used your 60 messages for this hour. The hour starts with your first message. Wait the number of minutes the message shows. Asking many questions within one minute hits a shorter limit, and the error looks the same; then wait a minute. When no wait time is known, the error says to try again later.
The assistant is temporarily unavailable. The assistant is busy or the model is not responding. Try again in a few minutes.
This project is not available. The project was deleted, or you are no longer a member. The message offers two buttons. Ask without the project switches to the Method and app panel, where you can ask again. Open projects closes the panel and takes you to your projects list.
Something went wrong. Any other error. Try again; if it keeps happening, try later.
The connection dropped. The part of the answer that arrived stays on screen. Pressing Stop does the same and marks the answer as stopped. Either way the partial answer is not sent with your next question. Under an error, Try again sends the question once more.
An orange warning under an answer means the assistant wrote numbers it could not match to the documentation or to your project's data. Do not rely on those numbers until you have found them on the result page.
What is kept¶
The conversation lives in your browser tab. It is not saved to your account or on the server, apart from the 14-day trace record that What the assistant sends describes. Closing the tab erases it, and so does signing out. Reloading the tab keeps it.
Each project has its own conversation, and the Method and app panel has a separate one. Clear conversation erases only the one that is open. A confirmation appears first, because this cannot be undone.
Your data¶
To answer, the assistant sends your question and the data of the open project to an AI model in the EU. That data includes other members' names, positions and opinions, since the project belongs to all of them. What the assistant sends lists it field by field, and the Privacy notice covers how long each part is kept.
If you do not want other members' assistant to show your opinions, turn on Hide my opinions from other members' assistant in your profile. It still shows them to you. The project result is calculated as before and still includes your opinion.
The switch has two limits. If no one else in the project has hidden theirs, your values can be worked out from the result and the other opinions. The assistant is told not to do that calculation. Other members may also notice that one opinion is missing.
Do not type sensitive data into the assistant, such as health information or other people's contact details.
Next: a worked example, where the four figures of a result come from a real panel of thirteen.