SQLite Storage

The SQLite backend provides a robust, file-based storage engine with ACID transactions, indexed queries, and built-in vector column support. It is the recommended backend for production deployments with up to 100,000 memories.

Configuration

PRX_MEMORY_BACKEND=sqlite
PRX_MEMORY_DB=./data/memory.db

The database file is created automatically on first run. All tables, indexes, and vector columns are initialized by PRX-Memory.

Schema Overview

The SQLite backend stores memories in a structured schema:

Column	Type	Description
id	TEXT	Unique memory identifier
text	TEXT	Memory content
scope	TEXT	Memory scope (global, project, etc.)
tags	TEXT	JSON array of tags
importance	REAL	Importance score (0.0--1.0)
created_at	TEXT	ISO 8601 timestamp
updated_at	TEXT	ISO 8601 timestamp
embedding	BLOB	Vector embedding (if enabled)
metadata	TEXT	Additional JSON metadata

Vector Storage

When embedding is enabled, vector data is stored as BLOB columns in the same table as the memory entry. This co-location simplifies queries and avoids join overhead.

Vector similarity search uses brute-force cosine similarity computation over the stored vectors. For datasets under 100,000 entries, this provides sub-second query times (p95 under 123ms based on benchmarks).

Maintenance

Compaction

Over time, deletions and updates can leave fragmented space. Use memory_compact to reclaim space:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "memory_compact",
    "arguments": {}
  }
}

Backup

The SQLite database file can be backed up by simply copying the file while the server is stopped:

cp ./data/memory.db ./data/memory.db.backup

WARNING

Do not copy the database file while the server is running. SQLite uses write-ahead logging (WAL) and a file copy during writes may produce a corrupt backup. Stop the server first or use the memory_export tool for a safe export.

Migration from JSON

To migrate from the JSON backend to SQLite:

1. Export your memories using memory_export.
2. Change the backend configuration to SQLite.
3. Import the exported data using memory_import.

Or use the memory_migrate tool for a direct migration.

Next Steps

• Vector Search -- How similarity search works internally
• Storage Overview -- Compare all backends
• Configuration Reference -- All environment variables