# AI Memory

The AI remembers durable facts your team states in chat and references them in every conversation on the property.

AI Memory lets the AI keep durable facts about a property between conversations. When someone on your team states something worth keeping - your fiscal year, a campaign naming rule, the KPI that actually matters - the AI saves it and references it in every future chat on that property.

Memories are **shared by the whole team**: one memory pool per property, so everyone gets consistent answers regardless of who taught the AI the fact.

## How Memories Get Saved

The AI saves memories on its own when you state durable facts in conversation. You can also be explicit:

- **"Remember that our fiscal year starts in April."** - saves a new memory
- **"Actually, our fiscal year starts in May - update that."** - updates the existing memory instead of creating a duplicate
- **"Forget what I told you about the fiscal year."** - deletes it

When a memory is saved during a chat, a **Memory updated** chip appears in the conversation header, so nothing is stored without you noticing.

The AI only saves facts *you* state. It will not save numbers from analytics data, content fetched from your connected tools, or sensitive personal information.

## How Memories Get Used

Every conversation on the property starts with the AI already aware of the saved memories - no lookup needed, no repeating yourself. Each memory carries the date it was last stated or confirmed, so when facts change over time, newer information wins.

If a property accumulates more memories than fit comfortably in a conversation, the most recent ones are included directly and the AI searches the rest when relevant (semantic search when your team has an OpenAI provider configured, keyword search otherwise - the same mechanism as the [Knowledge Base](/docs/knowledge-base/)).

## Managing Memories

Navigate to any property and click **AI Memory** in the sidebar. The Memories page lists every saved fact with its date and who saved it. From there you can:

- **Add** a memory manually
- **Edit** anything that's wrong or outdated
- **Delete** individual memories - deletions take effect from the very next message
- **Clear all** memories for the property

Memories are independent of conversations: deleting a conversation does not delete the memories saved from it, and deleting a memory removes it from all future conversations immediately.

## Incognito Conversations

Sometimes you want an answer uninfluenced by saved context - a fresh take, or an exploration you'd rather not have remembered. Toggle **Incognito** in any conversation's header (the ghost icon, visible to the conversation owner) and that conversation neither reads nor saves memories. Toggle it off again at any time.

## Turning Memory Off

If a property shouldn't use memory at all, switch it off under **Property Settings → AI Memory**. Existing memories are kept (and remain manageable from the Memories page) but are no longer referenced or written by conversations.

## Limits

| Plan | Memories per property |
|------|----------------------|
| Free | 20 |
| Pro | Unlimited |
| Business | Unlimited |

When a free property reaches its limit, the AI can still update existing memories - it just can't add new ones until you delete some or upgrade.

## Memory vs. Knowledge Base

Both give the AI context about your property, and they complement each other:

- **[Knowledge Base](/docs/knowledge-base/)** - documentation you write deliberately: detailed markdown entries about data quirks, definitions, and seasonal patterns. Best for anything that needs explanation or history.
- **AI Memory** - short facts captured automatically in the flow of conversation. Best for things you'd otherwise repeat: definitions, preferences, dates, naming rules.

A good rule of thumb: if it needs a paragraph, it belongs in the knowledge base. If it fits in a sentence, memory will catch it on its own.