Limitations

dbswitch is intentionally small. Know what it does not do before you adopt it — for anything beyond this surface, use your driver directly. dbswitch is not meant to hide SQL you actually need.

• Results are map[string]any. Value types are whatever the driver returns — a Postgres UUID comes back as [16]byte, a timestamp as time.Time. Convert at your boundary. There is no struct mapping.
• Conditions are equality-only, ANDed. WHERE col = val AND …. No <, >, OR, IN, LIKE, or joins yet. (Sorting, limits, and both cursor and offset pagination are available via List — the cursor is the one comparison you get.)
• SELECT * only — you can't yet choose which columns are returned.
• CreateTable is CREATE TABLE IF NOT EXISTS, not migrations. No schema versioning, alters, or indexes beyond column constraints. Use a real migration tool for evolving schemas.
• No transactions API.
• DynamoDB filters on non-"id" fields cost a full-table Scan. There's no secondary index by default, so Find/List/Count/Update/Delete with a non-key condition walk the whole table. Fine for small tables and convenience use; add a Global Secondary Index and query it directly with the AWS SDK for anything at scale.
• DynamoDB only supports a single "id" partition key — no composite (partition + sort) keys, and Unique is only enforceable on that key.

If your needs outgrow this, drop down to your driver directly (e.g. jackc/pgx or aws-sdk-go-v2) — dbswitch composes fine alongside raw driver usage.