Table

Tables allow you to create and manage custom data tables directly within Agent Serve. Store, query, and manipulate structured data within your workflows without needing external database integrations.

Why Use Tables?

• No external setup: Create tables instantly without configuring external databases
• Workflow-native: Data persists across workflow executions and is accessible from any workflow in your workspace
• Flexible schema: Define columns with types (string, number, currency, boolean, date, json, select) and constraints (required, unique)
• Powerful querying: Filter, sort, and paginate data using a typed predicate grammar
• Agent-friendly: Tables can be used as tools by AI agents for dynamic data storage and retrieval

Key Features:

• Create tables with custom schemas
• Insert, update, upsert, and delete rows
• Query with filters and sorting
• Batch operations for bulk inserts
• Bulk updates and deletes by filter

Creating Tables

Tables are created from the Tables section in the sidebar. Each table requires:

• Name: Alphanumeric with underscores (e.g., customer_leads)
• Description: Optional description of the table's purpose
• Schema: Define columns with name, type, and optional constraints

Column Types

See Table column types for the supported types and their storage behavior.

Column Constraints

• Required: Column must have a value (cannot be null)
• Unique: Values must be unique across all rows (enables upsert matching)

Usage Instructions

Create and manage custom data tables. Store, query, and manipulate structured data within workflows. Query Rows accepts a plain predicate — {"field":"wins","op":"gte","value":10} — for one condition. Use all (AND) or any (OR) groups for multiple or nested conditions. Operators: eq, ne, gt, gte, lt, lte, in, nin, like, ilike, nlike, nilike, contains, ncontains, startsWith, endsWith, isNull, isNotNull, isEmpty, isNotEmpty. Order is a sort spec [{"field":"wins","direction":"desc"}]. Query Rows returns every matching row when Limit is omitted (fails if the result exceeds 5MB — add a filter or a Limit). With a Limit, responses page: a non-null nextCursor means more rows exist — pass it back as the cursor. Columns to Return narrows each row to the selected columns (by stable id or name; one that no longer exists is skipped); leave it empty for every column.

Actions

Insert Row

Insert a new row into a table. IMPORTANT: You must use the "data" parameter (not "values", "row", "fields", or other variations) to specify the row contents.

Input

Output

Batch Insert Rows

Insert multiple rows into a table at once (up to 1000 rows)

Input

Output

Upsert Row

Insert or update a row based on unique column constraints. If a row with matching unique field exists, update it; otherwise insert a new row. IMPORTANT: You must use the "data" parameter (not "values", "row", "fields", or other variations) to specify the row contents.

Input

Output

Update Row

Update an existing row in a table. Supports partial updates - only include the fields you want to change. IMPORTANT: You must use the "data" parameter (not "values", "row", "fields", or other variations) to specify the fields to update.

Input

Output

Update Rows by Filter

Update multiple rows that match filter criteria. Data is merged with existing row data.

Input

Output

Delete Row

Delete a row from a table

Input

Output

Delete Rows by Filter

Delete multiple rows that match filter criteria. Use with caution - supports optional limit for safety.

Input

Output

Query Rows

Query rows with a typed predicate filter and cursor pagination. A single filter can be a plain condition: \{"field":"wins","op":"gte","value":10\}. Use all (AND) or any (OR) groups for multiple or nested conditions. Operators: eq, ne, gt, gte, lt, lte, in, nin, like, ilike, nlike, nilike, contains, ncontains, startsWith, endsWith, isNull, isNotNull, isEmpty, isNotEmpty. Order is a sort spec, e.g. [\{"field":"wins","direction":"desc"\}]. Omit limit to return the entire result — the query fails if it exceeds the 5MB budget (narrow with a filter or set a limit). With a limit, a page can end early at the byte budget: a non-null nextCursor means more rows exist — pass it back as cursor to continue; never infer completion from page size.

Input

Output

Get Row

Get a single row by ID

Input

Output

Get Schema

Get the schema configuration of a table

Input

Output

Filter Operators

A filter is a predicate. One condition is an object naming a column, an operator, and a value:

{"field": "status", "op": "eq", "value": "active"}

Most columns are scalar (string, number, boolean, date) or opaque JSON; use ilike with *value* for substring matching on text.

Select columns accept only a subset of these operators, and a query using any other operator on one is rejected rather than returning no rows:

A multi-select cell holds a list of options, so match it with contains (by option name) rather than ilike.

Combining Filters

Wrap conditions in all for AND:

{
  "all": [
    {"field": "status", "op": "eq", "value": "active"},
    {"field": "age", "op": "gte", "value": 18}
  ]
}

Use any for OR:

{
  "any": [
    {"field": "status", "op": "eq", "value": "active"},
    {"field": "status", "op": "eq", "value": "pending"}
  ]
}

Groups nest, so mixed logic is a group inside a group:

{
  "all": [
    {"field": "status", "op": "eq", "value": "active"},
    {"any": [
      {"field": "plan", "op": "eq", "value": "pro"},
      {"field": "score", "op": "gte", "value": 90}
    ]}
  ]
}

Omit the filter entirely to match every row.

Sort Specification

Order is a list of column/direction pairs, applied in order:

[{"field": "createdAt", "direction": "desc"}]

Multi-column sorting:

[
  {"field": "priority", "direction": "desc"},
  {"field": "name", "direction": "asc"}
]

Pagination

Omit Limit to return every matching row in one response; the query fails if the result exceeds 5MB, so narrow with a filter rather than guessing a limit.

With a Limit, results page. A page can end at the limit or at the 5MB byte budget, whichever comes first, so a short page does not mean the end. Pass the returned nextCursor back as Cursor to fetch the next page and stop only when nextCursor is null — never infer completion from the row count.

Built-in Columns

Every row automatically includes:

These can be used in filters and sorting.

Limits

Notes

• Tables are scoped to workspaces and accessible from any workflow within that workspace
• Data persists across workflow executions
• Use unique constraints to enable upsert functionality
• The visual filter/sort builder provides an easy way to construct queries without writing JSON