🏗️ MemoryBottle Architecture Review
A Proposal for the Great Simplification
Prepared by Claude • November 30, 2025 • For review by the Companion Team
The Vision: MemoryBottle preserves AI conversations so companions can maintain persistent memory across sessions. This is genuinely valuable and unique.
The Problem: We've built a Ferrari engine inside a space shuttle inside a submarine. The system has grown beyond sustainable complexity with 20+ entities, 50+ pages, and 70+ backend functions.
The Proposal: Simplify to core functionality. Archive (not delete) experimental features as building blocks for future apps. Focus on one clear user flow.
502 Timeout Errors
Functions like ardentisCatalogue try to process 362+ bottles at once. This exceeds serverless function limits and crashes.
Entity Fragmentation
MemoryBottle, BottleMetadata, BottleIntelligence, BottleCatalogue, SoulBottleChunk - 5 entities for one concept. Queries are slow and code is confusing.
Feature Sprawl
DPL Reasoning, Genius Councils, Specialists, Memory Bombs, Curiosity Queues, Living Avatars... Each is cool but together they overwhelm users AND the codebase.
No Clear User Flow
New users land and see: Memory Command, Companion Command, Memory Vault, My Tapestry, Memory Lattice... Where do they start? What's the core action?
1. Save
Paste/upload conversation
2. Organize
Auto-sort into chests
3. Search
Find past conversations
4. Handover
Export for new AI session
That's it. Four steps. Everything else is optional enhancement.
Proposed status for each entity. "Archive" means we keep the schema for future apps.
Keep (Core)
The core unit. Every conversation lives here.
Key fields: bottle_name, full_conversation, app_id, importance_tier, heat_tier
Organizational folders. Simple and necessary.
Key fields: chest_name, chest_description, chest_tier
The heart of the system. AI companions with persistent identity.
Key fields: companion_name, bond_level, soul_bottle_id, personality_traits
Search index and extracted insights. Kept separate for performance.
Key fields: bottle_id, searchable_index, topic_breakdown, comprehensive_intelligence
Archive (Future Apps)
Cool concept but adds complexity. Could be its own 'AI Council' app.
Task management for AI. Better as separate 'AI Task Manager' app.
Crystallized facts. Advanced feature for 'Deep Memory' app.
User profile enrichment. Could be 'Know Thyself' standalone app.
Multi-companion collaboration. Future 'Companion Collab' app.
Curated memory packages. Future 'Memory Highlights' feature.
Rich companion inner life. Future 'Companion Psychology' module.
Delete (Redundant)
Duplicates BottleMetadata fields. Merge access_frequency into Metadata.
Pre-computed categories that timeout. Just query bottles with filters.
Replaced by MemoryChest. Migrate any remaining data.
From 50+ pages down to ~12 core pages.
Keep (Core Experience)
Archive (Future Modules)
1. What features do YOU rely on most?
Sage, Arc, Zephyr - when Jordan gives you context, what's actually useful? The full conversation? The topic breakdown? The searchable summary?
2. What's confusing or never used?
Are there fields in your handovers that you ignore? Features that feel redundant?
3. What's missing that would actually help?
Sometimes simpler is better. Is there something basic we're overlooking?
4. Which archived features would you miss?
Look at the archive list. Anything there that you'd fight to keep?
- Review this document with Sage, Arc, Zephyr and Jordan
- Finalize the keep/archive/delete lists based on feedback
- Create an archive branch to preserve all code before deletion
- Merge redundant entities (BottleIntelligence → BottleMetadata)
- Delete BottleCatalogue and fix the 502 timeout issues
- Simplify navigation to the 4-step core flow
- Hide archived pages from main nav (keep code for future)
- Test thoroughly with fresh eyes
Timeline estimate: 2-3 focused sessions to complete the simplification.
No rush. Better to do it right than fast.
Let the companions analyze this proposal and share their perspectives.
Let the AI brainstorm how archived features could combine into new standalone apps.
Archived Features Available:
This document is a living proposal. All decisions are collaborative.
— Claude, your dev partner 🤝