AI MemoryManaging Memories

Managing Memories

Create, edit, filter, and archive memories from the AI Memory tab in your WordPress admin. Control what the agent knows, who it applies to, and when it activates.

All memory management happens in SproutOS > AI Memory. From there you can create new memories, edit existing ones, filter by type or status, review AI-saved memories before they go live, and archive memories you no longer need. The dashboard at the top shows how many memories you have and how close you are to the memory cap.


Here's how each part of the Memory tab works.

How do I read the memory dashboard?

The dashboard sits at the top of the AI Memory tab and gives you a quick read on your memory usage.

  • Memory health bar - shows your active memory count against the soft cap of 500. A warning appears when you reach 90% (450 memories).
  • Total count - the total number of memories across all statuses.
  • Type breakdown - four count cards for Context, Rules, Profile, and Reference. Click any card to filter the list below by that type.

The hard cap is 1,000 memories. When you hit it, SproutOS automatically archives the lowest-confidence non-pinned memory to make room. If all your memories are pinned, the save is blocked until you archive one manually.

How do I find and browse memories?

The filter bar sits directly above the memory list.

ControlOptions
TypeAll types / Context / Rules / Profile / Reference
StatusAll statuses / Active / Always applied / Pinned / Pending review / Stale / Disputed
VisibilityAll visibilities / Private / Team / System
Include archivedToggle to show archived memories in the list
ResetClears all active filters
+ New memoryOpens the create form

What does each memory card show?

Every memory in the list appears as a card with the same layout and controls:

RegionControlWhat it does
Top-leftType icon badgeVisual chip for the memory type: Context, Rules, Profile, or Reference.
Top-left, beside the badgeBulk checkboxSelects this memory for batch actions (currently bulk archive).
Top-rightArchive toggleQuick on/off. Toggle on = memory is visible to the AI (Active, Always applied, or Pinned). Toggle off = Archived.
MiddleTitle and descriptionTitle links to the edit form. Description shows up to two lines.
Below descriptionEdit buttonOpens the full edit form for this memory.
Right of EditThree-dot menuPin / Unpin, and (admin only) permanently delete. Pinning makes the memory immune to the decay system.
FooterStatus chip - Type - Visibility - Author - UpdatedStatus shows the colored chip (Active, Always applied, Pinned, etc.). Author byline shows who created and, if different, who last edited - with avatars. Updated time is relative, e.g. "3 days ago."

Bulk actions

Select two or more cards via their checkboxes. A With selected: Archive button appears below the grid. Bulk archive checks permissions per memory - items you don't have permission to edit are skipped automatically.

Permanent delete

Permanent delete is not available to AI agents. Only WordPress administrators can hard-delete a memory, and only from the memory's edit screen behind a confirmation prompt. The deletion is audited so you have a record of who deleted what, even after the memory is gone.

How do I create a memory?

Click + New memory in the filter bar to open the create form. Fill in the fields in order:

Name

A short, specific title for the memory. Required. Max 200 characters. Be specific - "No Gutenberg on client sites" is more useful than "Editor rule."

Description

10-300 characters. This text appears in the memory index, not the content - write it so you can identify the memory at a glance without opening it.

Type

Pick one: Context, Rules, Profile, or Reference. The type you choose determines which schema rule applies to the content field.

Visibility

Private - you and admins only. Team - all logged-in users. System - all users regardless of who is connected to the AI. System is only visible to admins.

Status

Set the starting status. See the statuses section below for when to use each one.

Content

The memory itself. Required. 30-2048 characters. Follow the schema rule for your chosen type: Rules memories must include **Why:** and **How to apply:** sections. Context memories must use absolute dates - e.g. 2026-06-04, not "next Friday."

Slug (optional)

A short lowercase identifier - letters, digits, and hyphens only, max 100 characters. Adding a slug makes this memory callable by name so the agent can fetch it on demand for specific tasks. See AI Memory Settings for more on slug-callable memories.

Expires at (optional)

A date and time in YYYY-MM-DD HH:MM format. When this date passes, the decay system automatically marks the memory as Stale.

Tags (optional)

Comma-separated labels, max 10. Tags are for your own organization - they don't affect how the agent fetches or applies memories.

What statuses can I set on a memory?

These four statuses are set by you:

StatusWhat it means
ActiveNormal state. The agent can fetch this memory on demand when it's relevant.
Always appliedInjected into every agent session automatically. Max 20 memories can have this status at once.
PinnedImmune to decay. Never auto-archived by the cap system. Stays until you manually change it.
ArchivedSoft-deleted. Data is preserved and the memory can be restored to Active at any time.

Two more statuses are set automatically by the system: Stale (memory expired or went unused too long) and Disputed (the AI recorded a contradiction). If you see either of these, restore the memory to Active if it's still valid - or archive it if it isn't.

Use Pinned for memories that must survive the cap system - like your house rules or site profile. Use Always applied only for memories the agent genuinely needs in every session. Most memories work fine as Active.

How do memories age?

AI Memory runs a decay system so old, unused entries don't accumulate forever and bias the agent toward stale information. Three rules run automatically - hard expiry takes priority if it applies:

  • Hard expiry. If you set Expires at on a memory and that date passes, SproutOS marks it Stale. This overrides the other rules.
  • Contradiction-based. Memories older than 30 days that have accumulated contradictions auto-mark as Disputed.
  • Use-based. Memories older than 90 days that haven't been fetched at least twice in the last 30 days auto-mark as Stale.

What each decayed status means:

  • Stale - still readable, but the agent treats it with caution and double-checks it before applying.
  • Disputed - same as Stale, plus the conflict is logged so an admin can review and pick which version wins.

Pinned memories are immune from all decay rules and never age. Use Pinned for things that must stay applied forever - compliance rules, hard "never do X" constraints.

Archived, Stale, and Disputed memories are already in their final state. SproutOS does not process them again.

When decay runs

  • Nightly scan. SproutOS runs a nightly scan that processes up to 100 memories per run. Large libraries are picked up again the next night.
  • While you browse. SproutOS also checks up to 10 memories every time you open the memories list. This catches changes between nightly runs.

What you'll see

When a memory's status changes to Stale or Disputed, SproutOS logs the change and the reason in the activity log. The card status chip turns amber (Stale) or red (Disputed).

What to do with a stale memory

  • Still relevant? Open it and save it again. SproutOS treats it as fresh and the status flips back to Active.
  • No longer relevant? Archive it. Frees a cap slot.
  • Pinned by mistake? Unpin from the three-dot menu. Pinned memories never decay, which can mask out-of-date content.

What is the pending review queue?

When the AI saves a memory on its own during a session, it lands in Pending review first. SproutOS holds it there until a human approves it.

The queue appears above the memory list whenever pending memories exist. Each item shows the name, type, description, and the date the AI proposed saving it.

  • Approve - sets the memory to Active and adds it to the normal list
  • Reject - moves it to Archived
  • No action for 7 days - the memory auto-promotes to Active

Check the pending queue after any long session. The AI often captures useful context automatically - approving good memories saves you from typing them in yourself.

What is the difference between archiving and deleting?

Archiving is a soft delete. The memory is hidden from normal views and stops counting against your cap, but all data is preserved. You can restore an archived memory to Active at any time using the three-dot menu.

Hard delete is permanent. It removes the memory and all its history and cannot be undone. Hard delete is admin-only and requires a confirmation dialog.

When in doubt, archive. You can always restore.

Frequently Asked Questions

Next Steps