9-step execution loop. The LLM fires only at Step 5.
Identifies user → loads Ware
Classifies task → selects Lobe(s) + tier
BDI: beliefs → desires → intentions → plan
LOBE.md + Glaze + Character + Memory + Blackboard
This is the ONLY step that calls the LLM. LLM = 20%. Runtime = 80%.
The ONLY step that calls the LLM. Runtime = 80%.
Blackboard receives Lobe outputs. Vault handles credentials.
Confidence scoring + entity extraction + bitemporal + contradiction check
Mark completed intentions, check replanning triggers
Return to user