Memory Triage & RAG System

Memfit AI's intelligence is enhanced by sophisticated memory management and knowledge retrieval systems.

Memory Triage (Intelligent Hippocampus)

Memory Triage acts as the system's intelligent hippocampus, managing long-term memory persistence and retrieval.

C.O.R.E. P.A.C.T. Framework

Memory fragments are assessed and scored across multiple dimensions:

Dimension	Description	Weight
Connectivity	Links to other memories	High
Origin	Source reliability	Medium
Relevance	Task applicability	High
Emotion	User sentiment signals	Low
Preference	User preferences	Medium
Actionability	Practical utility	High
Completeness	Information wholeness	Medium
Temporality	Time relevance	Medium

Memory Lifecycle

Input → Scoring → Threshold Check → Persistence
                       ↓
              Below Threshold → Discard
                       ↓
              Above Threshold → Index → Vector DB

Potential Questions Index

High-score memories are indexed with:

• Predicted future queries
• Semantic associations
• Contextual tags
• Usage patterns

RAG System (External Brain)

The RAG (Retrieval-Augmented Generation) system serves as an active, agentic knowledge service.

Hybrid Indexing

The RAG system uses multiple indexing strategies:

Vector Indexing

• Semantic similarity search
• Embedding-based retrieval
• Fuzzy matching

Keyword Indexing

• Exact term matching
• Technical terminology
• Identifier search

Retrieval Capabilities

Scalar Filtering

• Filter by metadata
• Time-based filtering
• Source filtering
• Type filtering

Multi-hop Retrieval

• Follow knowledge chains
• Aggregate related information
• Build comprehensive context

Knowledge Sources

The RAG system fuses multiple knowledge sources:

┌─────────────────────────────────────────────┐
│              RAG Knowledge Base              │
├─────────────────────────────────────────────┤
│  ┌─────────────┐  ┌─────────────────────┐   │
│  │   Domain    │  │    Tools & Forges   │   │
│  │  Knowledge  │  │    Documentation    │   │
│  └─────────────┘  └─────────────────────┘   │
│  ┌─────────────┐  ┌─────────────────────┐   │
│  │  Historical │  │     External        │   │
│  │   Memories  │  │    Knowledge        │   │
│  └─────────────┘  └─────────────────────┘   │
└─────────────────────────────────────────────┘

Integration with Execution

Context Enhancement

Before each decision:

1. Query formulation based on current state
2. Memory retrieval for relevant history
3. Knowledge retrieval for domain expertise
4. Context assembly for LLM

Active Knowledge Service

The RAG system is not passive:

• Proactively suggests relevant information
• Updates knowledge during execution
• Learns from successful interactions

Configuration

Memory Settings

Setting	Description	Default
scoreThreshold	Minimum score for persistence	0.6
maxMemories	Maximum stored memories	10000
decayRate	Time-based relevance decay	0.1

RAG Settings

Setting	Description	Default
topK	Number of results to retrieve	5
minSimilarity	Minimum similarity threshold	0.7
multiHopDepth	Maximum retrieval depth	2