AI Memory - How It Works
AI Memory stores your WordPress site rules, context, and resources so the agent walks into every session already knowing your site - no repeating yourself.
Every time you start a new AI session, you normally re-explain your WordPress site's setup, rules, and preferences from scratch. AI Memory eliminates that. It stores your working conventions, current site context, brand profile, and external resource links directly in WordPress and injects them automatically at the start of every session.
You control what gets remembered, who can see each memory, and when it applies.
Why AI Memory exists
AI agents forget everything between sessions. The usual fix - a textarea where you type "act as a senior dev who uses Bricks" - has no audit trail, no expiry, no way for a teammate to see what got saved or who saved it.
AI Memory replaces that pattern with a structured audit trail. Every save writes a log entry with a fingerprint of the content - not the content itself. Months later you can click any memory and see exactly which conversation produced it, who saved it, and what abilities the agent had active at the time.
The memory content never lands in the audit log in plain text - only the fingerprint.
That receipt is the difference between "AI took notes" and "AI followed rules a human can audit." That's why SproutOS built this for production sites, not just dev sandboxes.
Three tools in one. AI Memory replaces what other tools split into three separate concepts: Always applied memories are custom instructions, slug memories are skills, and the rest are long-term memory. One system, one place to audit.
Memory Types
4 types cover every kind of context - project state, team rules, brand profile, and external links.
Visibility Levels
Private, Team, or System - control who the agent applies each memory for.
Seed Memories
4 blank starters ship with SproutOS. They cost zero tokens until you fill them in.
On/Off Toggle
Disable AI Memory site-wide at any time. Your data is preserved when it's off.
Here's how each part of AI Memory works.
What are the 4 memory types?
Every memory has a type. The type tells SproutOS what kind of context it holds and enforces a schema rule to keep the content structured and useful to the agent.
| Type | Purpose | Schema Rule |
|---|---|---|
| Context | Project state, current goals, deadlines, in-progress work | Must use absolute dates - e.g. 2026-06-04. Never use relative dates like "next week" or "Thursday" |
| Rules | Team conventions, design constraints, working preferences | Must include both a **Why:** line and a **How to apply:** line |
| Profile | Audience, brand voice, business goals, personal preferences | No extra schema requirement. Defaults to Private visibility |
| Reference | Links to external tools, docs, and systems | Must include a URL or a recognizable identifier - e.g. Notion:, Linear:, Figma:, GitHub: |
Following the schema rules makes memories more useful, not just valid. A Rule without a Why: is easy to misapply. A Context memory with a relative date becomes meaningless after a week.
How do visibility levels work?
Every memory has a visibility level that controls who the agent applies it for:
- Private: Only the WordPress user who created the memory and site administrators. Agents connected as a different user won't see it.
- Team: Any logged-in user on the site. The default for most memories. Suits agency teams where every team member's AI sessions should share the same playbook.
- System: Applies to every agent connection on this site, regardless of which logged-in user the agent is connected as. Only administrators can set this, and only administrators can edit or archive a System memory. Still requires authenticated access - System visibility does not expose memories to anonymous visitors.
What are seed memories?
When you first activate SproutOS, four blank memories are created automatically. These are the seed memories:
| Name | Type |
|---|---|
| Project context | Context |
| House rules | Rules |
| Site profile | Profile |
| External resources | Reference |
Seed memories ship with Always applied status and System visibility. The moment you fill one in, every AI session on the site picks it up automatically.
Until you add content, seed memories cost zero tokens. SproutOS skips empty memories entirely during injection - they're placeholders waiting for you.
Start with House rules. Add your one or two most important working conventions - like "always ask before deleting anything" or "use Elementor, not Gutenberg." That single memory will shape every session from day one.
How do I turn AI Memory on or off?
The AI Memory toggle appears at the top of the SproutOS > AI Memory tab. It is admin-only.
When you turn AI Memory off:
- No AI Memory abilities register with your AI client
- No memories are injected into agent sessions
- All existing memory data is preserved and available when you turn it back on
Turning AI Memory off is a clean pause. Nothing is deleted.
Frequently Asked Questions
Nothing breaks. Empty seed memories are skipped entirely and cost zero tokens. They sit there until you're ready to fill them in. You can also delete them if you prefer to start from scratch.
Editors can create Private and Team memories. System memories - the ones that apply to all users regardless of who is connected - are admin-only. The System visibility option only appears in the create form for admin users.
No. Three gates protect you:
- The agent must declare one of five allowed trigger types for every save - an explicit user request, a repeated correction, a confirmed unusual choice, a stated fact, or an inferred fallback.
- Every save goes through a type-specific schema check. Rules memories require Why and How to apply sections. Context memories reject relative dates. Reference memories need a URL or identifier.
- A near-duplicate check blocks saves that are 90% or more similar to an existing memory. The agent is told to update the existing one instead.
Below-threshold saves land in the pending review queue for your approval. There's a soft cap at 500 and a hard cap at 1,000. When you hit the hard cap, SproutOS auto-archives the lowest-confidence non-pinned memory to make room.
There's a master toggle in the header of the AI Memory tab. Flip it off and:
- Memory abilities stop being exposed to connected agents
- Nothing gets injected into the agent system prompt
- The nightly decay scan stops running
- The pending review auto-promote timer stops running
Your existing memories stay in the database untouched until you flip the toggle back on.
Yes. AI Memory is builder-agnostic - it works on any WordPress site regardless of which builder or e-commerce stack you're running. SproutOS ships separate abilities for WooCommerce and Elementor. AI Memory is the layer underneath that lets agents track preferences and rules across all of them.