🏗️ Architecture: How should Vitto's memory system be architected to achieve J...

Date: 2026-03-19
Time: 21:55:53 UTC
Session: gspace1-239.2
Agent: backend
Confidence: 94%
Status: ✅ Documented

📋 Decision Context

Question:

How should Vitto's memory system be architected to achieve Jarvis-level agent capabilities with human-like recall, context persistence, and multi-scope knowledge management?

Task: S239: Vitto Memory System V2 — Jarvis-level memory architecture

🎯 Decision Made

Chosen:

Two-system architecture: Soul (identity/consciousness — WHO Vitto is) separated from Brain (memory scopes — WHAT Vitto knows). Brain uses 4-scope hierarchy: Organization → Space → Universe → User. Three intelligence innovations: (1) Admission Control with 5-factor scoring gate, (2) Prospective Indexing that pre-generates future questions at write time, (3) Dopamine-Gated Dual Route for fast/slow recall. Cognitive plugin system provides lifecycle hooks for extensibility.

Options Considered:

1. Single flat memory store with tag-based filtering
2. 3-tier hot/warm/cold with no scope awareness
3. Full neuroscience brain simulation (frontal/parietal/temporal lobe services)
4. Soul+Brain dual system with 4-scope hierarchy and cognitive plugins

💭 Reasoning

Deep analysis of 5 existing memory layers (VittoMemory, UserDossier, AgentSession, IdentitySynthesis, SpaceIntelligence) revealed: (a) No scope separation — all memories were flat, making multi-tenant impossible. (b) Token budget was inverted — system context got priority over user context. (c) In-session memory loss due to missing conversation history injection (fixed S237) and no virtual context management. (d) No admission control — every interaction stored, causing noise. (e) No conflict resolution between competing memory scopes. Research into state-of-art systems (MemGPT, Kumiho/D-MEM, Zep) showed prospective indexing achieves 93.3% recall accuracy vs 70-80% for pure vector search. The Soul/Brain separation prevents conflating identity customization with knowledge storage — architecturally cleaner and maps to existing UI structure (MIND vs IDENTITY sidebar groups). 4-scope hierarchy maps directly to the platform's tenant model (org → space → universe → user).

📁 Files Involved

• APIs/backend-vitto-api/src/app/mind/memory/vitto-memory.types.ts
• APIs/backend-vitto-api/src/app/mind/services/dopamine-router.service.ts
• APIs/backend-vitto-api/src/app/mind/services/memory-admission.service.ts
• APIs/backend-vitto-api/src/app/mind/conflict/memory-conflict-resolver.service.ts
• APIs/backend-vitto-api/src/app/mind/plugins/cognitive-plugin.types.ts

🔗 Related

• Decision ID: dev_1773957353689_rccoy4k
• Session: 239
• Workspace: gspace1

Auto-generated by MCP Smart Consolidation