Row-level reads and writes

Turso holds application data in libSQL tables that mirror SQLite. When a Beam agent receives a trigger, such as a new record arriving upstream, it reads the relevant rows, applies the rule your team approved, and writes the update back to the table where connector permissions allow. Straightforward inserts and edits run untouched. Rows that conflict with an existing key, fail a validation you defined, or would overwrite data marked as protected are held and passed to a person, who confirms the change before the agent commits it to the database.

Schema change detection

Turso tables carry a defined schema of columns and types. A Beam agent watches for structural differences between an incoming payload and the current table, such as a field that no longer maps to a column. For expected, pre-approved additions it records the row and notes the variance. For anything larger, a renamed column, a dropped field, or a type it cannot safely cast, the agent stops and routes the case to an engineer with the before and after laid out, so a human decides whether the schema itself should change.

Scheduled table backups

Turso keeps production data your team cannot afford to lose. On a schedule you set, a Beam agent reads a table or a defined query result and copies it to the destination your connector allows, then records that the snapshot completed. If a copy returns fewer rows than the last attempt, times out, or the destination rejects the write, the agent does not silently retry forever. It notifies the owner with the counts from both attempts, and a person decides whether to rerun the backup or investigate the source.