Phase 4D: Implement query interface with Knowledge Graph context

[timkicker]

Goal

Natural language queries resolved against Knowledge Graph using two-step model pipeline.

Query pipeline

1. User natural language query
2. Model call 1: generate Cypher query
3. Query validator: known node types, no write ops, LIMIT required
4. Execute against Graph Daemon
5. Model call 2: format results as natural language (separate call from step 2)
6. Relative time expressions resolved to absolute UTC before generation

Validator rules

• Only known node types and relations
• No MERGE, CREATE, DELETE, SET operations
• Must contain LIMIT clause
• On failure: feed error back for retry
• After 2 failed attempts: inform user, no silent fallback

Project context when Focus Mode active

• Context seed: project name, description, tags, accounts, recent activity
• context_scope = 'project': AI read scope restricted to PART_OF edges of active project

Waypointer AI mode

• Tab toggle in Waypointer activates AI mode
• Results as structured typed results in same interface

References

Blueprint Sections 5.5, 5.6