Contact Memory
How Goldilocks remembers and uses information about returning contacts
Contact Memory allows your AI agent to remember information about contacts across conversations. This enables personalized support and contextual responses.
How Memory Works
- Contact identified by ID (passed from your system)
- AI extracts relevant information during conversations
- Facts stored in memory categories
- Future conversations access stored memories
- AI uses memories to personalize responses
Memory Categories
Categories define what types of information to remember.
Viewing Categories
- Go to AI Agents > Memory
- See list of memory categories
- Each shows name, description, and status
Default Categories
Goldilocks includes common categories:
| Category | Description |
|---|---|
| Preferences | How contact prefers to interact |
| Known Issues | Past problems they've experienced |
| Account Info | Key account details mentioned |
| Product Usage | How they use your product |
Creating Categories
- Go to AI Agents > Memory
- Click Add Category
- Fill in:
- Name - Display name
- Slug - Unique identifier
- Description - What to extract
- Extraction Prompt - Optional custom prompt
- Save
Example Category
Name: Communication Preferences
Description:
How the contact prefers to communicate and be addressed.
Includes preferred name, communication style, and contact preferences.Extraction Prompt:
Extract any stated preferences about how the contact wants to be
addressed or communicated with. Look for preferred names, titles,
or explicit preferences about communication style.Memory Extraction
Automatic Extraction
After conversations, the AI analyzes the exchange and extracts relevant facts into memory categories.
Example conversation:
Contact: "Please call me Mike, not Michael"
Extracted memory: Category: Communication Preferences Value: "Prefers to be called Mike instead of Michael"
What Gets Extracted
The AI looks for:
- Explicit preferences stated by contact
- Important facts about their situation
- Context that would help future interactions
- Corrections or updates to previous information
Extraction Timing
Memory extraction happens:
- When a conversation ends
- After significant exchanges
- When relevant information is detected
Using Memories
In Conversations
Memories are automatically included when:
- Contact is identified
- Memories exist for that contact
- Memories are relevant to current topic
Example:
Memory: Prefers email contact, had billing issue in March
AI response: "Hi Mike! I see you had a billing question. Since you prefer email, I can send you a detailed breakdown there. Would you like that?"
In Workflows
Memory conditions can trigger workflows:
# Trigger VIP workflow for high-value customers
conditions:
- category: customer_tier
operator: equals
value: "vip"Contact Identification
For memory to work, contacts must be identified.
How to Identify Contacts
Pass contact ID in your widget configuration:
new Goldilocks({
apiKey: 'your-key',
customerId: 'unique-customer-id',
customerLabel: 'John Smith' // Optional display name
});Without Identification
If no contact ID is passed:
- Conversation happens normally
- No memories are retrieved
- No new memories are stored
- Each conversation is independent
Managing Memories
View Contact Memories
- Go to AI Agents > Memory
- Find contact section or search
- View all memories for a contact
Edit Memories
- Find the memory entry
- Click to edit
- Update the value
- Save
Delete Memories
- Find the memory entry
- Click delete
- Confirm
Delete memories when contacts request it or when information becomes outdated.
Privacy Considerations
Data Retention
Configure how long memories are kept:
- Set expiration per category
- Automatic deletion after period
- Manual retention management
Contact Requests
Be prepared to:
- Show contacts what's remembered
- Delete memories on request
- Correct inaccurate memories
Sensitive Data
Be careful with:
- Personal health information
- Financial details
- Sensitive personal data
Configure categories to avoid extracting sensitive information if inappropriate.
Best Practices
Focus on Helpful Information
Remember things that improve support:
- Preferences that personalize experience
- Context that avoids repetition
- History that informs responses
Don't Over-Remember
Avoid storing:
- Trivial details
- Temporary information
- Potentially sensitive data
Review Regularly
Periodically check:
- Are memories accurate?
- Are categories useful?
- Any outdated information?
Test Memory Flow
- Create a test contact
- Have a conversation mentioning preferences
- Start a new conversation
- Verify memories are used appropriately
Persona Scoping
Limit memory categories to specific personas:
- Edit the memory category
- Find Persona Scope
- Choose Selected and pick personas
- Save
Use this when different personas need different memory types.