Memories

What your agents are learning about users: volume, type breakdown, and where the knowledge gaps are.

When an agent has memory enabled, it stores facts, preferences, and instructions across conversations so it can personalize replies. The Memories page in the organization view is where the analysis layer aggregates that store across the whole fleet, turning what individual agents learn into a single, governed picture of what your organization knows about its users.

Period selector

The date filter in the top-right scopes every chart and KPI on the page: presets of 7 days, 30 days (the default) or 90 days, or a custom range of UTC days. The page keeps its own window, independent of the dashboard’s. Knowledge gaps are cumulative and do not depend on it.

A refresh button next to it forces a re-fetch of the analytics.

KPIs

Card	What it means
Total Memories	All memories captured in the period. Trend badge is the percentage change against the previous period of the same length.
Daily Average	Average new memories captured per day.
Knowledge Gaps	Total topics where users asked something the agent didn’t know yet. The high-priority count is shown below.
Affected Users	Estimated number of distinct users who hit at least one knowledge gap.

Daily new memories chart

A bar chart showing how many memories were captured each day in the period. Use it to spot bursts (a feature launch, an incident) and lulls (an outage, a holiday).

Type breakdown

Memories are tagged with a type by the capture pipeline:

Type	What it stores
Facts	Verifiable information about a user: role, location, customer ID, domain.
Preferences	What the user likes, dislikes, prefers: communication style, language, format.
Relationships	How the user relates to people, teams, or systems: manager, team members, key accounts.
Instructions	Standing orders the user has given: “always summarize in three bullets”, “never use Markdown”.

The breakdown shows count and percentage per type, plotted as horizontal bars.

Knowledge gaps

A gap is a topic users asked about repeatedly that the agent doesn’t yet have a memory for. Each gap card shows:

• The topic name and a severity badge (high, medium, low).
• The number of occurrences and the number of distinct users affected.
• Up to five extracted keywords.
• A suggested action, usually “add a knowledge base entry covering X” or “extend the agent’s instructions with Y”.

Use the gaps list as a backlog: each high-severity gap is a concrete documentation, knowledge base, or instruction change to ship.

Recommendations

A standalone card at the bottom lists priority-ordered actions the analytics pipeline derived from the gaps and the type distribution. Each entry has a priority badge (high / medium / low) and a short action string. These are advisory; the platform doesn’t apply them automatically.

Where to go next

See memories grouped by topic

Drill into specific topics, examples, and trend direction.

Read user adoption signals

Personalization depth as part of the adoption picture.