Storage Layer

Memory Systems

Tiered memory architecture enabling agents to store, retrieve, and learn from experience across multiple storage backends.

Three-Tier Architecture

In-Memory Store

Fast local storage for immediate context and current conversation state.

Sub-millisecond access
Session-scoped
Message history
Current context

Use case: Active conversation context, immediate decisions

Distributed Memory (Redis)

Cross-agent shared state for ACMF modes, learned patterns, and coordination.

Cross-agent access
State persistence
Pattern sharing
Pub/sub messaging

Use case: ACMF state, agent coordination, shared patterns

Vector Memory (Weaviate)

Semantic search across knowledge base using embedding vectors.

768-dim embeddings
Similarity search
Knowledge base
Batch operations

Use case: Semantic retrieval, pattern matching, knowledge lookup

Memory Types

Short-term

Message history, current conversation context

Retention: Session

Long-term

Pattern library, learned best practices

Retention: Permanent

Episodic

Historical deployment outcomes, incident records

Retention: 90-day decay

Semantic

Embedding-based knowledge for similarity search

Retention: Versioned

Vector Database Configuration

vector_db:
  provider: weaviate
  weaviate:
    host: localhost:8080
    scheme: http
    connection_pool:
      max_idle_conns: 10
      max_open_conns: 100

embedding:
  provider: sentence-transformers
  model: all-MiniLM-L6-v2
  dimension: 384

search:
  default_limit: 10
  max_limit: 100
  min_certainty: 0.7

knowledge:
  decay_half_life: 90  # days
  relevance_formula: (confidence × decay × 0.5) + (success_rate × 0.3) + (use_count × 0.2)

Knowledge Graph Architecture Overview