Knowledge Layer
Nova has two memory layers plus one execution resolver. They are separated on purpose.
Why separation matters
Natural language can be fuzzy, but execution must be reliable.
- memory helps Nova understand context
- deterministic resolution decides who/what to execute against
The three parts
- Contact Graph: people and communication identity knowledge
- Life Graph: user-owned preferences/routines/places
- Action Resolution: deterministic target selection for execution
Contact Graph (people memory)
Examples:
- "Karol is my mom"
- alias and relationship hints for contact resolution
Use case:
- supports "email my mom" style requests safely
Life Graph (personal context)
Examples:
- "No meetings before 9"
- "My favorite restaurant is Uchi"
- named places like home or office
Use case:
- planning context and better suggestions
Action Resolution (execution safety)
Action resolution turns natural references into concrete, safe targets using canonical contacts and approved/high-confidence knowledge.
Data flow (simple)
User text
-> extraction
-> policy checks
-> contact/life memory updates
-> planning context query
-> deterministic action resolution