Skip to main content
Most flows start by fetching a list — rows from a sheet, records from a CRM, items from an API — and the shape you get is rarely the shape you want. The Data section of the palette reshapes it before it reaches whatever comes next. Every Data block takes a list in and produces Result (the new list) plus Count, so they chain end to end. A list can be a previous step’s output, JSON text, or a single object.

​
The blocks

Map Items

Keep only the fields you name, under the names you want. The biggest lever on payload size.

Filter Items

Keep only the items matching a condition, with an optional second condition.

Group & Aggregate

Collapse many rows into one item per group — count them, total them, merge their values.

Unique Items

Remove repeats, keeping the first of each.

Sort Items

Order a list by a field, ascending or descending.

Take Items

Take the first N items, optionally after skipping some.

Build Array

Create a list from scratch, or append to one you already have.

Parse JSON

Turn JSON text into a list you can map, filter and group.

To JSON Text

Turn a list back into JSON text for a response or message body.

Run JavaScript

A short script, for anything the blocks above cannot express.
Field names are matched without regard to capitalisation, and a name with dots in it reaches inside nested values — customer.city reads the city inside customer.

​
Map Items

Give it a Field mapping: a JSON object of the name you want on the left, the field it comes from on the right.
Everything not listed is dropped. Turn on Keep unmapped fields to keep the rest as well, in which case a mapping simply renames.

​
Filter Items

Pick a Field, a Condition and a Value. Available conditions: equals, does not equal, contains, does not contain, starts with, ends with, greater than, greater than or equal, less than, less than or equal, is empty, is not empty, is one of, is not one of. The last two take a comma-separated list. Add a Second field to test two things at once, combined with AND (both must match) or OR (either may match).
Comparisons like greater than compare numbers as numbers when both sides look numeric, so 10 is correctly greater than 9. You do not have to declare a column type.

​
Group & Aggregate

This is the block for turning a repeated export into one row per thing. A schedule with one row per doctor per day becomes one row per doctor. Set Group by to the field (or comma-separated fields) that identify a group, then list what each group should contain in Aggregations:
Each entry needs a name (the output field) and an op. Every op except count needs a field. Include group fields carries the group-by fields onto each result item. Leave it on unless you have a reason not to.
When a field varies inside a group, first throws the rest away. A doctor who works at three clinics has three different Locations in the group; first keeps one and silently loses two. Use join_distinct (or list_distinct) for any field that legitimately differs within a group. The same applies to Unique Items — it keeps one whole item and discards the others.
If you leave Aggregations empty and turn Include group fields off, there is nothing to put in each group. The step will tell you rather than quietly returning empty items.

​
Filter before you group

Rows that should not count must be removed before grouping. In a staff schedule, a day someone is off is still a row, and it still carries a location — group without filtering and you will credit people with places they never work.
1

Filter Items

Time Start does not equal Off
2

Group & Aggregate

Group by Doctor Name, merging Location with join_distinct
3

Map Items

Keep just the fields the answer needs
4

To JSON Text

Compact, into the response body

​
Run JavaScript

For a transform the other blocks cannot express. Prefer a regular block where one fits — those show their inputs and outputs in the run history, while a script is opaque when something goes wrong.
  • Read an earlier step with getOutputFrom('<step id>', '<Output>'). The Use a previous step picker adds the reference to the end of your script.
  • Rows behave like objects: r['Doctor Name'] and r.Department both work.
  • End with a plain expression — out; — because the result is the value of the last expression. A script that ends with a loop returns that loop’s last value instead, which is rarely what you meant.
Outputs are Result (the value), Count (items, when it is a list) and Text (the value as JSON).
Scripts run in a sandbox. They cannot reach the server, the filesystem or the network, and they are limited in how long they may run and how much memory they may use — a script that never finishes fails its step instead of holding up anything else.

​
Feeding an AI agent tool

When a flow answers an AI agent tool, what the agent receives is capped at 24,576 characters. Beyond that the response is cut short, and the agent is told it is incomplete so it can ask again more narrowly — but a shortened answer is still a worse answer. The fix is to send less, not to send more:

Send only what is needed

Map Items to a handful of fields. Dropping the columns a question never asks about is the single biggest saving.

Split by question

One tool returning a directory, another taking filters and returning detail, beats one tool trying to return everything.

Stay compact

Leave Indented off in To JSON Text — formatting is wasted space for a machine reader.

Watch the size

To JSON Text reports Length, so you can branch when a payload grows.

​
Next steps

Building a flow

The canvas, the palette, connecting steps and publishing.

Runs and troubleshooting

See what a step actually produced when a flow misbehaves.