People & Profiles

The profiles system gives agents persistent memory about people in the user’s life. Stored facts can be retrieved in later interactions, subject to user scope and tool/context availability.

How It Works

When a user mentions personal details (“my mother lives in Berlin”, “my boss prefers email”), agents can store those facts in a profile. On a later request, the agent can retrieve the stored city; it still needs a confirmed delivery address and authorization before placing an order.

The user’s own profile is included in supported agent prompt construction:

USER CONTEXT:
Name: Max
- location: Berlin, Germany
- timezone: Europe/Berlin
- language: English, German

Profile Structure

Field	Description
name	Display name (“Mom”, “Dr. Mueller”, “Acme Corp”)
relationship	Relation to user (self, mother, colleague, boss, friend)
category	Entity type: person, organization, pet
facts	Key-value facts with source and timestamp
isUserProfile	If true, injected into agent system prompts

Facts

Each fact has:

• key — location, birthday, email, phone, likes, dislikes, occupation, timezone
• value — the actual information
• source — how it was learned (user told us, learned from conversation)
• learnedAt — timestamp

Agent Tools

Available when the profiles tool is exposed to the current role and permitted:

Tool	Description
list_profiles	List all profiles for the current user
get_profile	Get a profile by ID or name
create_profile	Create a new profile
update_profile	Update profile fields
add_fact	Add or update a fact
remove_fact	Remove a fact by key
delete_profile	Delete a profile
search_profiles	Search by name or fact values

Use Cases

• Morning briefings — agent knows your timezone and location for weather
• Scheduling — agent knows your colleagues’ preferences
• Gift giving — agent knows what your family members like
• Communication — agent knows preferred contact methods for people