The Memory Service stores messages, meetings, web sources, entities, relationships, profile items, and source capsules in a private long-term memory layer.
- Multi-channel recall across semantic, keyword, graph, and time signals
- Salience scoring, decay, consolidation, reflection, and evidence retention
- Per-user memory boundaries and explicit scope handling
Save meaningful sources with low friction while keeping ordinary browsing quiet.
- Save selected text, whole pages, tables, SVG/visual regions, and Jira owner comments
- Store source, reason, anchors, takeaways, future triggers, and undoable events
- Use suggestion and auto-save gates to avoid turning browsing history into memory noise
Bring relevant existing memories into the page, message, issue, meeting, or selected text you are looking at now.
- Quiet context recall with rest, hover peek, and expanded card states
- Selection Memory Search for user-initiated text lookup
- Source links, why chips, feedback, site mute, and privacy-sensitive blocking
Use memory to draft insertable text beside native input boxes without taking over the send action.
- Works with RingCentral replies, Jira comments, and Web AI input boxes
- Creates reply drafts or context packs with preview and confirm gates
- Learns writing style from redacted edit traces instead of storing final messages
Turn the day and each meeting into a small set of actionable, evidence-backed cues.
- Today missions from meetings, reminders, actions, relationship signals, and project risks
- Meeting prep handoff into RingCentral meeting capture and side panel
- Action items, decisions, timelines, summaries, and post-meeting panorama
Ask your own memory questions and inspect the evidence trail behind answers.
- Topic locking for short questions such as "is that BE ready?"
- Evidence-backed answers with unknowns, prior answers, and external-check handoff when needed
- Search results, timeline, source memory detail, action queue, and decision center
Organize long-running work and people context without pretending every derived signal is a confirmed fact.
- Project dashboard with local plans, evidence coverage, risk cues, and Memory Service watched-project sync
- Relationship Radar for recent interactions, open loops, context cards, and meeting briefs
- Safe evidence links and confirm/reject paths for sensitive or inferred relationship facts
See what Personal AI can access, where coverage is stale, and what needs explicit user control.
- Memory Coverage Map with platform health, quality score, repair queue, and import drawer
- Backup, restore, privacy controls, source-level links, and site-level mute/block lists
- Human confirmation only for high-responsibility boundaries such as external sending, irreversible deletion, or stable profile facts