Remind + SQLite

Use an explicit SQLite store when you need a SQL database file. Remind's default durable store is JSON under .remind/memory; it needs no database server.

Prerequisites

• Python 3.11+
• Remind installed:

pip install remind-mcp

Quickstart

1. Configure Remind

Copy the example config from the repository’s examples/sqlite folder to either the global or project-local location:

# Global (applies to all projects)
mkdir -p ~/.remind
cp remind.config.json.example ~/.remind/remind.config.json

# Project-local (only this directory, takes precedence over global)
mkdir -p .remind
cp remind.config.json.example .remind/remind.config.json

Edit the file and fill in your API keys.

2. Use Remind

Select SQLite explicitly for this example:

export REMIND_STORE="sqlite:////absolute/path/to/project/.remind/remind.db"

Without that setting, Remind uses the nearest parent .remind/memory store, or creates one in the current directory.

# Store a memory
remind remember "First memory stored in SQLite"

# Recall memories
remind recall "memory"

# Check stats
remind stats

# Run consolidation
remind consolidate

3. Inspect the database directly

sqlite3 .remind/remind.db

SELECT COUNT(*) FROM episodes;
SELECT id, title FROM concepts;
.tables  -- list all tables
.quit

Choosing a database file

Default local store — stores JSON records alongside your project:

unset REMIND_STORE
remind remember "something"
# → .remind/memory in the current directory on first use

Named store — stores JSON in ~/.remind/<name>.memory:

remind --db myproject remember "something"

Explicit path or URL — full control:

remind --db /path/to/my.db remember "something"
remind --db "sqlite:////path/to/my.db" remember "something"

Environment variable (recommended for shared scripts):

export REMIND_DB_URL="sqlite:////path/to/my.db"
remind remember "something"

Config file (~/.remind/remind.config.json):

{
  "db_url": "sqlite:////path/to/my.db"
}

Database URL format

sqlite:////absolute/path/to/file.db
sqlite:///relative/path/to/file.db

SQLite requires four slashes (////) for absolute paths and three (///) for relative paths. See the SQLAlchemy docs for details.

Vector search (sqlite-vec)

Remind can store embedding vectors in sqlite-vec vec0 tables for fast KNN search. The sqlite-vec Python package is installed with remind-mcp, but the extension only activates if your Python was built so that sqlite3.Connection supports enable_load_extension (i.e. SQLite was compiled with loadable extensions, and Python was configured accordingly). On some macOS and pyenv setups this is off by default.

Quick check:

python -c "import sqlite3; c=sqlite3.connect(':memory:'); print('sqlite-vec usable:', hasattr(c, 'enable_load_extension'))"

If that prints False, Remind still runs and recalls correctly using in-Python cosine similarity; only the native index path is skipped. To enable sqlite-vec, reinstall Python linked against Homebrew SQLite with --enable-loadable-sqlite-extensions (see Configuration — Vector search).

When to switch to PostgreSQL

SQLite is great for single-user and local development use cases. Consider switching to PostgreSQL when you need:

• Multiple processes or machines writing concurrently
• A shared team database
• Better performance at very large scale

See the postgres-docker example on GitHub to get started with PostgreSQL.