Metadata backends¶

A backend stores graph definitions, snapshots, tags, and storage partition metadata. The backend is metadata-only; artifact bytes still live in the configured storage system.

Memory¶

The default backend is in-process memory. It is useful for examples and tests, but it does not persist snapshots across processes.

SQLite¶

SQLite is the recommended local durable backend:

from pathlib import Path

from arti.backends.sqlite import SQLiteBackend

backend = SQLiteBackend(path=Path("artigraph.sqlite"))

Use SQLite for local examples, notebooks, and demos where you want tags and snapshots to survive process restarts without running a service.

Postgres¶

Postgres is intended for shared durable metadata:

from arti.backends.postgres import PostgresBackend

backend = PostgresBackend.from_url(
    "postgresql+psycopg://artigraph:password@localhost/artigraph",
)

Both SQL backends apply bundled Alembic migrations on first use by default. Set auto_upgrade=False if you want to control migrations explicitly.

Trust boundary¶

Persisted metadata includes module-qualified Python type paths, so artifact and producer definitions must live at module scope in an importable module. Definitions in a script run as __main__ cannot be persisted. Restoring graph metadata imports the recorded modules, so treat the backend as trusted infrastructure.