Data Manipulation

Data manipulation statements change table contents. In v0.1, the core workflow is straightforward: insert rows, update rows by predicate, delete rows by predicate, and vacuum storage after deletes when compaction matters.

Use column-targeted inserts for application code. The statement says exactly which columns receive values.

INSERT INTO Users (Id, Name, Score, IsActive)
VALUES (1, 'Ada', 98.5, true);

Full-row inserts are shorter, but they depend on table column order.

INSERT INTO Users
VALUES (2, 'Grace', 95.0, true);

Insert vectors as bracketed numeric literals.

INSERT INTO Embeddings (Id, Label, Emb)
VALUES (1, 'red', '[1,0,0]');

INSERT INTO Embeddings (Id, Label, Emb)
VALUES (2, 'warm-red', '[0.9,0.1,0]');

Update rows with SET and WHERE.

UPDATE Users
SET Score = 99.0
WHERE Id = 1;

Update more than one column in the same statement.

UPDATE Users
SET Score = 100.0,
    IsActive = false
WHERE Name = 'Ada';

Delete rows with a predicate.

DELETE FROM Users
WHERE IsActive = false;

After many deletes, use VACUUM to compact the table.

VACUUM Users;

Wrap related writes in a transaction when they should commit or roll back together.

BEGIN TRANSACTION;

INSERT INTO Users (Id, Name, Score, IsActive)
VALUES (3, 'Katherine', 97.0, true);

UPDATE Users
SET Score = 98.0
WHERE Id = 3;

COMMIT;

Rollback discards buffered changes in the current session.

BEGIN TRANSACTION;

DELETE FROM Users
WHERE Id = 3;

ROLLBACK;

DML Support Summary

Supported DML: column-targeted and full-row INSERT (the value count must match the columns), vector literal inserts (for example '[1,0,0]'), UPDATE ... SET ... WHERE ... (with a fixed-width compiled fast path for narrow source-generated single-row shapes), DELETE ... WHERE ..., VACUUM table, and session-bound transactional DML (BEGIN TRANSACTION, COMMIT, ROLLBACK). MERGE and the PostgreSQL COPY protocol are not supported in v0.1. See the SQL compatibility matrix for the authoritative feature status.