rAlph – What rAlph learns about you
rAlph is only available when AI is enabled for the tenant.
So that rAlph understands your tenant better and better over time, it keeps a knowledge model: a growing collection of short, durable statements about your organization (e.g. “internally we call objectives north stars”) and about individual users’ preferences (e.g. “keep answers concise”).
This knowledge is stored in natural language and is independent of the AI model: switch the provider or model and everything learned is preserved.
Where rAlph’s knowledge comes from
Section titled “Where rAlph’s knowledge comes from”There are three sources:
- Automatically from conversations. After a few chat turns, rAlph distils durable knowledge from the transcript in the background. Such entries start as candidates and must be confirmed (see Admin review) before they feed into future answers.
- Explicitly with
/remember. Tell rAlph in chat with the/remembercommand (German/merken) what to remember. rAlph decides whether it is team knowledge (for everyone in the tenant – admins only) or a private preference (just for you), and briefly confirms what it stored. - Through your feedback. When an answer used learned knowledge, a small bar appears beneath it: thumbs up strengthens the knowledge, thumbs down weakens it, and Correct lets you enter the right version – which then supersedes the wrong entry.
Team knowledge vs. private preference
Section titled “Team knowledge vs. private preference”- Team knowledge (tenant scope) applies to everyone in the tenant and may only be created or confirmed by administrators.
- Private preferences (user scope) belong to you alone. They feed only into your conversations and do not appear in anyone else’s admin review.
Trust, decay, retire
Section titled “Trust, decay, retire”Every entry has a confidence. It rises when the same knowledge is corroborated repeatedly or gets a thumbs up, and falls on a thumbs down. Knowledge left unused for a long time slowly decays and eventually stops being used. Only confirmed knowledge with sufficient confidence feeds into answers – and live tool data always takes precedence over memory.
Admin review
Section titled “Admin review”Under AI settings → Learned knowledge, administrators see the team-wide knowledge. There you can confirm candidates, edit entries, or retire them. Private user preferences are not shown here, for privacy.