Memory & AI insights
Memory is the “context layer” that helps the AI behave consistently over time. It reduces repeated questions and helps operators understand what has already been learned.
At a glance
- Memory lives on the contact detail page.
- It includes:
- Summary
- Open items
- Structured AI insights (facts)
- Interaction log
- Rebuild from recent calls
- Memory can show LLM usage metrics (tokens/cost).
Memory works best when it is owned. Decide who can edit it and what counts as “truth” (AI-derived vs operator-confirmed).
What each section means
LLM usage
Shows model/provider usage and cost signals for memory extraction workflows.
Summary
A human-readable snapshot of the contact context.
Open items
Things that are unresolved or require follow-up.
AI Insights (facts)
Structured facts extracted from interactions. Insights can have provenance and may support:
- inline edits
- reset-to-AI behavior (when an operator wants to revert a manual override)
Interaction log
A history of memory-related events and recent interactions.
Rebuild from recent calls
Rebuild re-extracts memory from recent calls. It is powerful, but it can have cost and governance implications.
Rebuild can trigger additional extraction and cost. Use it when you have new high-signal calls or you fixed a bad prompt, not as a default habit.
When memory matters most
- High-touch accounts where context continuity affects conversion.
- Multi-touch sequences where follow-ups depend on prior calls.
- Ops teams that want predictable outcomes and fewer “AI surprises.”