Skip to main content
The AI gets better at working with you in two ways: it remembers what you tell it, so you don’t repeat yourself, and it records how your team rates its answers, so you can see where it falls short. This page covers both, and how to review or delete what is kept.

​
Memory

Ankra’s AI keeps a small personal memory: a learned profile (how you like answers delivered) and saved memories (facts worth keeping across conversations). Review both under AI → Settings → Memory.
Memory is personal. It only shapes your own conversations - nobody else in your organisation can read your profile or memories, and organisation admins have no special access to them.

​
Learned profile

As you chat, the AI picks up preferences - communication style, preferred level of detail, other notes - and applies them to future answers. The Learning switch pauses that: turn it off and the profile stays as it is until you turn learning back on. Reset profile wipes the learned traits and starts over. This can’t be undone.

​
Saved memories

Memories are short facts saved during conversations - either because you asked the AI to remember something or because it judged a detail worth keeping. In chat, the AI can also recall them on demand (the save_memory and search_memories MCP tools expose the same store to connected editors). From the settings page you can delete any single memory or Clear all. Deletion is immediate and permanent.

​
Scope

Memory is scoped per user, per organisation: what the AI learns about you in one organisation does not follow you into another.

​
Feedback analytics

Every AI response can be rated with a thumbs up or down, optionally with a comment. AI → Settings → Usage & feedback turns those ratings into a signal for how the AI performs on your clusters. Unlike memory, feedback is organisation-wide: admins see every rating.

​
Overview tab

  • Totals, positive/negative counts, and positive percentage for the selected period (default: last 30 days)
  • Daily trend chart
  • Per-cluster breakdown showing where the AI performs well or struggles
  • Count of feedback entries that include comments

​
Feedback list tab

A paginated table of individual entries. Each row carries full context: the feedback type and comment, the cluster, the AI’s response, and the user question that prompted it. Filter by date range, cluster, feedback type, free-text search, or “with comments only”. Admins can Export the filtered list as CSV.

​
Who sees what

  • Organisation admins see all feedback across the organisation and can export CSV.
  • Members see only their own feedback entries and analytics.
  • One rating per AI message - submitting twice is rejected.

​
Leaving feedback

Rate any AI Assistant response with the thumbs buttons under the message. Ratings with a short comment (“wrong namespace”, “fixed it first try”) are the most useful signal.

​
Next

Teach the AI your organisation’s conventions, rather than one person’s, with AI Skills.