User Guide

Memory Health

Corpus-ops dashboard for the persistent agent memory corpus: counts by type, repo, and tool; stale-note detection; a credential-pattern policy scanner; and vector-based near-duplicate detection.

Open in app All docs

Hands-on workflow

1
Open Memory Health and read the Overview plates for active note count and the type, repo, and tool breakdowns.
2
Switch to Staleness, choose a note type and a day threshold, then archive notes that are no longer accurate.
3
Open Policy to review notes flagged for credential-like patterns; archive any that should not be in searchable memory.
4
Use Duplicates to run a vector similarity scan and archive the weaker note in each near-duplicate pair.

Screen-by-screen

Cropped captures of each tab, taken from a sanitized test workspace.

Overview: active-note count plus breakdowns by type, repo, and tool.

Staleness: notes not updated within the chosen day threshold, oldest first, each archivable.

Policy: notes flagged by credential-pattern regex; matched secrets are never displayed inline.

Duplicates: pick a similarity threshold and run a vector scan to find near-duplicate notes.

Tabs and sub-tabs

The page may expose these sub-areas. Use this list as the quick orientation map before writing internal SOPs.

OverviewStalenessPolicyDuplicates

Tips and common mistakes

Staleness defaults to project notes at 90 days; widen the type filter to all types for a full sweep.
The policy scanner reports a hit count and excerpt but never prints the matched secret -- open the note to inspect.
The duplicate scan queries the vector index per note and can take 15-30 seconds; it is intentionally manual.
Archiving is reversible at the data layer: the row is retained for audit and only dropped from recall.