SQLite Memory

Zero-cost local memory system using SQLite. No API keys, no cloud calls, no vector embeddings. Just fast, reliable local storage.

Quick Start

from sqlite_memory import store_preference, store_decision, search_memory, get_memory_stats

# Store a preference
store_preference("User dislikes overly apologetic language", importance=1.0)

# Store a project decision
store_decision("Using pygame.mouse for head motion simulation", importance=0.9)

# Search memories
results = search_memory("sensitivity")
# Returns: [{id, category, content, importance, created_at}, ...]

# Get stats
stats = get_memory_stats()
# Returns: {total_memories, categories, db_path}

When to Use

Memory Categories

• preference — User likes/dislikes, style preferences (importance: 0.7-1.0)
• decision — Technical choices, design decisions (importance: 0.8-0.9)
• fact — General facts about user or project (importance: 0.5-0.7)
• project — Project-specific updates, milestones (importance: 0.7-0.9)
• correction — Explicit corrections from user (importance: 1.0)

API Reference

Storage Functions

store_preference(content: str, importance: float = 0.8) -> int
store_decision(content: str, context: dict = None, importance: float = 0.9) -> int
store_fact(content: str, importance: float = 0.6) -> int
store_project_update(project: str, content: str, importance: float = 0.7) -> int
store_correction(content: str, importance: float = 1.0) -> int

Retrieval Functions

search_memory(query: str, category: str = None, limit: int = 10) -> List[dict]
get_memory_stats() -> dict

Architecture

workspace/
└── memory.db              # SQLite database
    ├── memories           # Main memory table
    │   ├── id             # Auto-increment
    │   ├── category       # preference/decision/fact/project/correction
    │   ├── content        # The actual memory text
    │   ├── context        # Optional JSON metadata
    │   ├── importance     # 0.0-1.0, affects sort order
    │   ├── created_at     # Timestamp
    │   └── access_count   # Usage tracking
    └── sessions           # Session tracking (optional)

Comparison with Other Systems

Limitations

• Text search only — No semantic/vector search (use graph-memory if needed)
• Single user — No multi-user isolation
• No auto-extraction — Must manually call store functions (unlike Mem0)

Migration Path

If you outgrow sqlite-memory:

1. Need vector search? → graph-memory with Ollama (local embedding)
2. Need auto-extraction? → Mem0 + OpenAI API
3. Need cross-device sync? → elite-longterm-memory + SuperMemory cloud

Resources

• scripts/sqlite_memory.py — Core memory module, import and use directly
• references/usage.md — Detailed usage examples and patterns