Memory and Custom Instructions

Use Workspace instructions to set shared rules for your generated content. The Assistant also learns preferences from conversations. Explicit instructions take priority over learned preferences and default wording.

Auto-learning memory

As you chat with the assistant, it automatically picks up on important details and preferences from your conversations. For example:

• The areas you frequently create content about

• The tone you like your content written in

• Your favourite image styles and aspect ratios

• Common property types you deal with

• Client communication preferences

The assistant learns from patterns in your usage over time. For example, if you consistently choose "Abstract Growth" style images, it will start defaulting to that style. Preferences are confirmed after being observed at least three times, so a single choice does not lock anything in.

How it works

• Memories are extracted from your conversations automatically after you finish chatting

• They are classified by topic (e.g. "market knowledge", "style preference", "area focus")

• Usage patterns (image style, aspect ratio) are tracked as tallies that build over time

• Once the assistant has learned your usual aspect ratio, it flags it as "your usual" in the aspect ratio list when it asks. It always asks rather than choosing for you, so the choice stays yours

• When you start a new conversation, relevant memories are retrieved to provide context

• You do not need to repeat yourself, the assistant remembers

Viewing your learned preferences

Open Assistant memory to see what the assistant has learned about you. On a computer, click Menu at the bottom of the sidebar, then Settings, and choose Assistant memory under Account. On a phone, tap your initials in the top-right corner and find it under Account. What you will see includes:

• Your preferred image style (e.g. "Prefers Abstract Growth style images")

• Your preferred aspect ratio (e.g. "Usually uses 4:5 aspect ratio")

• Location preferences and other details picked up from conversations

You can delete any learned preference from this screen if it is wrong or outdated.

Telling the assistant to remember something

You can explicitly tell the assistant to remember a preference by saying things like "remember that I always want...", "from now on...", or "I prefer...". The assistant will confirm it has saved the preference immediately.

Privacy

• Conversation summaries remain personal to you within the workspace. They are not copied into pack or image prompts.

• Confirmed creative preferences can guide generated content across the workspace. Personal details, recipient preferences and one-off requests are excluded.

• Workspace instructions are visible to your team. The workspace owner or a platform admin manages them.

• They are stored securely in your workspace database

• Memory content is never used to train external AI models

Custom instructions

Workspace instructions guide Assistant writing, Content Packs, Property Packs, Swipes, property-feed posts, calendar captions, property-ad copy and AI images. They also apply when content is generated on a schedule.

For example: Use only postcode districts in captions and image text, rather than town names. Where a relevant postcode is known, generated image labels use it. Report data, factual property details and working URLs are preserved.

Instructions apply to new generation and requested edits. Existing posts, images and reports are not rewritten automatically. Reports keep their original data and place names.

Setting custom instructions

1. Open Assistant memory (on a computer, click Menu at the bottom of the sidebar, then Settings; on a phone, tap your initials. Either way, it is under Account)

2. As the workspace owner, turn on Workspace instructions

3. Write the creative rules your workspace should follow. If you are not sure where to start, tap What should I write? for some examples

4. Save

What to include

Good custom instructions might cover:

• Your role and agency: "I am a senior negotiator at Smith & Co in Didsbury, Manchester"

• Tone preferences: "Write in a friendly, professional tone. Avoid jargon"

• Content style: "Keep social media captions under 150 words. Always include a call to action"

• Local knowledge: "Our main areas are Didsbury, Chorlton, and Sale"

• Formatting: "Use bullet points for property features. Always mention the asking price"

Tips

• Be specific, vague instructions like "be helpful" do not change behaviour meaningfully

• Focus on preferences that apply to most conversations

• You can update your instructions at any time

• Custom instructions and auto-learned memories work together, instructions set the rules, memories provide the context

Toggling memory

Learned preferences on the Assistant memory page turns the auto-learning on and off. When it is off:

• The assistant will not extract or store new memories

• Your existing conversation memories are preserved but do not contribute to new content

• Workspace instructions, enabled creative preferences from other members and the content rules below remain separate

The What it picked up from your chats section contains conversation preferences. What it learned from your posts contains the preferences learned from your content decisions.

Content preferences learned from your posts

Content preferences is shared across your workspace. It learns from corrections in Swipes, Content Packs, Property Packs, the calendar, Library and approval queue. Paragraph changes count; mechanical corrections to facts, prices, spelling or links do not become a writing style.

The list holds at most 12 active preferences. Similar ideas are merged. A new preference replaces a weaker inference when appropriate; old preferences do not disappear just because they have been working for a long time. Silent approvals of your listing posts and silent Library saves are not style instructions.

Review, correct or remove a preference

1. Open Assistant memory from your account menu.
2. Find Content preferences and its active count.
3. Open Why it learned this to inspect supporting examples. Older preferences may not have a saved evidence trail.
4. Use the pencil to correct a preference. A corrected preference is marked Confirmed and will not be changed by automatic learning.
5. Use the X to remove a preference. It appears under Removed preferences, where you can deliberately restore it. Remove another active preference first if all 12 places are occupied.

Removal also prevents a paraphrase in that preference category being inferred again. Your explicit workspace instructions can still specify what you want.

Content preferences starts on. Use its on/off switch to pause automatic learning and use of inferred preferences. It preserves the list. Confirmed corrections and enabled explicit workspace instructions still apply. The conversation-memory switch is separate.

Read the Swipes guide.

Permanently delete a removed preference

Under Removed preferences, choose Delete permanently beside Restore, then confirm. The preference disappears from the list and cannot be restored. We retain a hidden block against learning it again; the deleted preference text and attached examples are removed. Your original post feedback remains separate.