Shree Bhanderi.

Homebase Building Block

Visible Memory.

A surface that lets the user inspect what the agent remembers, where it came from, when it was last used, and how to edit or remove it.

Read the thesis

Memory inspector

Memory should feel usable instead of creepy.

Calls server parsing with rate limits.

MemorySourceLast usedConfidenceUsed forControls
Prefers concise executive summariesEdited by userYesterdayHighWritingEdit / delete
Uses vendor scorecard formatPrior artifact8 days agoHighComparisonsEdit / delete
Avoids speculative claimsReview correction3 days agoMediumResearchEdit / delete
Weekly report goes to marketing leadCalendar/email2 weeks agoMediumReportingEdit / delete

Selected memory

Prefers concise executive summaries

Learned when the user edited a long research memo into a shorter recommendation format.

Learned

Explicitly saved

Examples

Used in vendor briefs and weekly updates.

Used in this answer

  1. Your vendor-analysis format
  2. The Q1 research export
  3. Your preference for concise recommendations

Memory becomes trustworthy when the user can inspect, correct, and delete it.

What it is

Visible Memory makes personalization inspectable. The user can see what the system knows, where it learned it, and why it used that context.

Problem

Persistent memory is powerful, but hidden memory creates distrust. Users need to see the system's assumptions before they can rely on it.

How it works

  • Track saved preferences, prior artifacts, reusable procedures, connected files, project history, and observed review patterns.
  • Show memory by source, recency, confidence, and use case.
  • Let users edit, delete, or scope memory before it influences work.

Why it matters

  • Trust requires inspectability.
  • Memory needs source, recency, and edit controls.
  • Personalization should feel useful rather than hidden.

Behavior

Good behavior

I used your prior vendor-analysis format, the Q1 research export, and the tone preference you saved last week.

Bad behavior

The system silently personalizes outputs without explaining what context shaped them.

Recruiter signal

This shows product judgment around trust, AI UX, systems thinking, and the difference between useful automation and opaque automation.