Skip to content
Stand With Ukraine

SQL expressions


When there is a need to use a piece of raw SQL, use the sql exported from the BaseTable file, it is also attached to query objects for convenience.

When selecting a custom SQL, specify a resulting type with <generic> syntax:

import { sql } from './baseTable';

const result: { num: number }[] = await{
  num: sql<number>`random() * 100`,

In a situation when you want the result to be parsed, such as when returning a timestamp that you want to be parsed into a Date object, provide a column type in such a way:

This example assumes that the timestamp column was overridden with asDate as shown in Override column types.

import { sql } from './baseTable';

const result: { timestamp: Date }[] = await{
  timestamp: sql`now()`.type((t) => t.timestamp()),

In some cases such as when using from, setting column type via callback allows for special where operations:

const subQuery ={
  sum: (q) => q.sql`$a + $b`.type((t) => t.decimal()).values({ a: 1, b: 2 }),

// `gt`, `gte`, `min`, `lt`, `lte`, `max` in `where`
// are allowed only for numeric columns:
const result = await db.$from(subQuery).where({ sum: { gte: 5 } });

Many query methods have a version suffixed with Sql, you can pass an SQL template literal directly to these methods. These methods are: whereSql, whereNotSql, orderSql, havingSql, fromSql, findBySql.

await db.table.whereSql`"someValue" = random() * 100`;

Interpolating values in template literals is completely safe:

// get value from user-provided params
const { value } = req.params;

// SQL injection is prevented by a library, this is safe:
await db.table.whereSql`column = ${value}`;

In the example above, TS cannot check if the table has column column, or if there are joined tables that have such column which will lead to error. Instead, use the column or ref to reference a column:

// ids will be prefixed with proper table names, no ambiguity:
db.table.join(db.otherTable, 'id', 'otherId').where`
  ${db.table.column('id')} = 1 AND
  ${db.otherTable.ref('id')} = 2

SQL can be passed with a simple string, it's important to note that this is not safe to interpolate values in it.

import { sql } from './baseTable';

// no interpolation is okay
await db.table.where(sql({ raw: 'column = random() * 100' }));

// get value from user-provided params
const { value } = req.params;

// this is NOT safe, SQL injection is possible:
await db.table.where(sql({ raw: `column = random() * ${value}` }));

To inject values into sql({ raw: '...' }) SQL strings, denote it with $ in the string and provide values object.

Use $$ to provide column or/and table name (column or ref are preferable). Column names will be quoted so don't quote them manually.

import { sql } from './baseTable';

// get value from user-provided params
const { value } = req.params;

// this is SAFE, SQL injection are prevented:
await db.table.where(
    raw: '$$column = random() * $value',
    values: {
      column: 'someTable.someColumn', // or simply 'column'
      one: value,
      two: 123,


import { sql } from './baseTable';

// simplest form:
sql`key = ${value}`;

// with resulting type:
sql<boolean>`key = ${value}`;

// with column type for select:
sql`key = ${value}`.type((t) => t.boolean());

// with column name via `column` method:
sql`${db.table.column('column')} = ${value}`;

// raw SQL string, not allowed to interpolate values:
sql({ raw: 'random()' });

// with resulting type and `raw` string:
sql<number>({ raw: 'random()' });

// with column name and a value in a `raw` string:
  raw: `$$column = $value`,
  values: { column: 'columnName', value: 123 },

// combine template literal, column type, and values:
sql`($one + $two) / $one`.type((t) => t.numeric()).values({ one: 1, two: 2 });


column references a table column, this can be used in raw SQL or when building a column expression. Only for referencing a column in the query's table. For referencing joined table's columns, see ref.

  // select `("table"."id" = 1 OR "table"."name" = 'name') AS "one"`,
  // returns a boolean
  one: (q) =>
    q.sql<boolean>`${q.column('id')} = ${1} OR ${q.column('name')} = ${'name'}`,

  // selects the same as above, but by building a query
  two: (q) => q.column('id').equals(1).or(q.column('name').equals('name')),


ref is similar to column, but it also allows to reference a column of joined table, and other dynamically defined columns.

await db.table.join('otherTable').select({
  // select `("otherTable"."id" = 1 OR "otherTable"."name" = 'name') AS "one"`,
  // returns a boolean
  one: (q) =>
    q.sql<boolean>`${q.ref('')} = ${1} OR ${q.ref(
    )} = ${'name'}`,

  // selects the same as above, but by building a query
  two: (q) =>


fn allows to call an arbitrary SQL function.

For example, calling sqrt function to get a square root from some numeric column:

const q = await{
  sqrt: (q) => q.fn<number>('sqrt', ['numericColumn']),

q.sqrt; // has type `number` just as provided

If this is an aggregate function, you can specify aggregation options (see Aggregate) via third parameter.

Use type method to specify a column type so that its operators such as lt and gt become available:

const q = await{
  // Produces `sqrt("numericColumn") > 5`
  sqrtIsGreaterThan5: (q) =>
      .fn('sqrt', ['numericColumn'])
      .type((t) => t.float())

// Return type is boolean | null
// todo: it should be just boolean if the column is not nullable, but for now it's always nullable