Memory Architecture

Prism uses a three-tier memory system that enables progressive personalization.

Overview

┌──────────────────────────────────────────┐
│            User Memory Space             │
│                                          │
│  ┌──────────────┐  ┌──────────────────┐ │
│  │  Short-Term   │  │    Long-Term     │ │
│  │  (Session)    │  │  (Persistent)    │ │
│  │               │  │                  │ │
│  │ Current chat  │  │ Trading style    │ │
│  │ Active trades │  │ Risk profile     │ │
│  │ Today's data  │  │ Token watchlist  │ │
│  │               │  │ Past performance │ │
│  └──────────────┘  └──────────────────┘ │
│                                          │
│  ┌──────────────────────────────────────┐│
│  │         Strategy Memory              ││
│  │                                      ││
│  │ Win/loss patterns per strategy       ││
│  │ Market regime performance            ││
│  │ Token-specific insights              ││
│  └──────────────────────────────────────┘│
└──────────────────────────────────────────┘

Short-Term Memory

Active during each session. Maintains:

• Current conversation context
• Open positions and intra-day events
• Enables coherent multi-turn dialogue

Long-Term Memory

Persists across sessions. Stores:

• Your trading preferences and risk tolerance
• Historical interaction patterns
• Token watchlist and favorites
• Cumulative performance data

Strategy Memory

Specialized performance tracking:

• Which strategies worked under which conditions
• Market regime classification (bull/bear/range) and corresponding performance
• Token-specific behavioral patterns observed over time