AI Memory
The AI assistant remembers facts, preferences, and context across conversations.
Contents
AI Memory
AI Memory allows the assistant to remember facts, preferences, and context across individual conversations. This means you don't have to repeat important information every time.
Prerequisite
AI Memory requires your own API key (BYOK). If you are using your organization's system key, this feature is not available. For more information, see the article AI Setup (BYOK).
How Memory Works
The assistant has three memory tools:
Remember
The assistant can remember information you share with it. This happens either at your request or automatically when the assistant recognizes relevant information.
Examples:
- "Remember: Our standard hourly rate is 120 euros"
- "Remember: We always use 19% VAT for clients in Germany"
- "Remember: Mr. Mueller prefers communication via email"
Recall
When you ask a question or describe a task, the assistant automatically searches its memory for relevant information and incorporates it into its response.
Forget
You can ask the assistant to forget specific information:
- "Forget what you know about our hourly rate"
- "Delete the memory about Mr. Mueller"
Memory Categories
Memories are automatically categorized:
| Category | Examples |
|---|---|
| Preference | Preferred language, default tax rate, formatting preferences |
| Fact | Hourly rates, bank details, company information |
| Context | Ongoing projects, current priorities |
| Instruction | Rules for invoice creation, naming conventions |
| Project | Project-specific details, deadlines, technologies |
| Client | Client preferences, contact persons, special requirements |
Automatic Memory
The assistant automatically remembers certain events without you having to ask:
- New client created -- Name and basic data are stored
- New project created -- Project name and type are noted
This keeps the assistant informed about your latest business developments.
Activity Learning
In addition to manual and automatic memory, the assistant analyzes your work activities every 6 hours and recognizes patterns:
- Most active projects: Where do you work the most?
- Working hours: When are you typically active?
- Task completion rate: How many tasks do you complete?
- Most frequent actions: What do you do the most?
- Weekly volume: How much do you accomplish per week?
These insights are stored as memories and help the assistant make better suggestions.
Memory Limits
- Maximum 200 memories per user
- The 20 most relevant memories are included in each conversation
- Relevance is determined by category, importance, and recency
- When the limit is reached, the oldest and least important memories are automatically replaced
Privacy and GDPR
You have full control over your stored data.
Viewing Memories
- Go to Settings > AI Memory
- You will see a list of all stored memories
- Filter by category, source, or search term
Deleting Individual Memories
Click the delete icon next to a memory to remove it individually.
Deleting All Memories
Click Delete All to remove all stored memories at once. This action cannot be undone.
What Is Stored?
- Only text information (no files or images)
- Memories are tied to your user account
- Other users in your organization cannot see your memories
- All memories are removed when your account is deleted
Practical Tips
Defining Rules
Share your business rules with the assistant so it considers them with every request:
- "Remember: Invoices to company XY always with 0% VAT (reverse charge)"
- "Remember: Always set project deadlines 2 weeks before the actual due date"
Storing Client Preferences
- "Remember: Client ABC always pays by bank transfer within 30 days"
- "Remember: For client DEF we use the hourly rate of 150 euros"
Cleaning Up
Regularly review your stored memories under Settings > AI Memory and delete outdated information. This keeps the assistant up to date and ensures the most relevant information is prioritized.
dVersum Tip
Use AI Memory strategically: store primarily recurring business rules and client preferences. This way the assistant doesn't need to be re-instructed with every request and can work faster and more precisely.
Last updated: 4/6/2026