Transforms

A transform is a small safe tool. It makes a new value and touches nothing else. It runs inside your server, with no call to the internet. It costs nothing. The same input always gives the same answer.

The settings of a transform sit flat on the step. They sit right next to id, transform, skip_when, and fail_on. There is no extra box around them. Use the setting names below exactly as written. A ? at the end of a name means you can leave that setting out. The ref {{item.*}} only works inside map expressions and inside foreach loops.

{ "id": "days", "transform": "date_diff",
  "from": "{{step.deal.last_activity_date}}", "to": "now", "unit": "days" }

#Collections

A collection is a list of items (an array). These tools work on lists.

#Reduce and map

reduce can use these operators: sum, count, avg, min, max, concat, or, and.

filter can take many conditions at once. By default all of them must be true (AND). Set mode: "any" and one true condition is enough (OR).

map has three kinds of expression:

• One ref. It pulls out that value and keeps its type.
• A math expression over {{item.*}}. It can use + - * / %, min, max, abs, ceil, floor, and round.
• An object template.

output_field adds the answer to each item. Without it, the answer replaces the item:

{ "id": "weighted", "transform": "map", "array": "{{step.deals}}",
  "expression": "{{item.amount}} * {{item.stage_probability}} / 100",
  "output_field": "weighted_amount" }

group_by with aggregate works out numbers for each group. It returns an array:

{ "id": "by_stage", "transform": "group_by",
  "array": "{{step.deals}}", "field": "stage",
  "aggregate": {
    "count": { "operator": "count" },
    "total": { "field": "amount", "operator": "sum" }
  } }

Without aggregate it returns { key: [items] }. To walk through the groups, chain to_list and then map.

#Objects

#Strings

#Encoding and size

These help in two cases. Some operations have a size limit. And sometimes you need an id that is always the same, so a re-run repeats the work instead of making a copy.

#Numbers

math can use add, subtract, multiply, divide, and modulo. It can also use abs, ceil, and floor. Those three take one number only.

#Dates

Units are days, hours, minutes, and seconds.

date_period takes one of these names:

today, yesterday, tomorrow, this_week, last_week, this_month, last_month, this_quarter, last_quarter, this_year, last_year, last_7_days, last_30_days, last_90_days, next_7_days, next_30_days.

The start day and the end day are both included. Weeks are ISO weeks (weeks start on Monday). The period comes from the run's own clock. So one run gets one fixed answer, and running it again gives that same answer. With date, you get the period around that time instead.

Give time_zone the value {{context.server.time_zone}}. That is your own time zone. Then a day starts at your midnight. Without it, a day starts at midnight in UTC (world clock time). In Los Angeles, that is 5 pm the day before. The time_zone setting works the same way in date_format and date_parse.

#Timestamps

#Calendar events

event_when takes a calendar event and says when it happens on your clock. It reads start_at, end_at, and is_all_day. It gives back start, end, day, time, date_time, and text. An all-day event says All day where a time would go. Use part to get just one of them, like "part": "time". Inside map, leave out field, and each whole event goes in.

#Logic

any and all evaluate only conditions. A condition written under values is text, and text is always true there, so the validator warns about it.

default treats an empty string as missing. It keeps 0 and false. Use it instead of coalesce when a template might turn into "".

#Privacy

Use hash_replace before an AI step. Use hash_restore after it. That way the AI model never sees the fields you chose. The mapping never leaves the run.

#Display

A checklist item looks like { label, issue, detail_ok, detail_issue, detail_null }. See Checklist rows for which one a row shows.

A table column can have one of these formats: text, currency, number, date, relative_date, boolean, percent.

A checklist item or a table column with type: "action" can carry an action. An action always shows as a button the owner clicks. The click opens a normal recipe run. All the usual rules and approvals still apply. An action never runs on its own.

to_csv escapes text the way RFC 4180 (the standard CSV rules) says. Pair it with a file write. Then you get a file a spreadsheet can open.

#Checklist rows

A checklist row's issue is a condition string. True shows as a problem (detail_issue). False shows as fine (detail_ok).

A row shows as unknown, with detail_null, when what it judges is missing:

• a comparison (equal, greater and the rest) whose value is missing;
• is_empty, is_null, is_not_empty or is_not_null on a field whose record is missing, such as {{step.contact.result.email}} when the contact was never read;
• is_not_empty or is_not_null on a whole step's output or a setting that is missing.

is_empty or is_null on a whole output or a setting that is missing shows the problem: the missing value is what it asks about. A row with no issue shows detail_ok, or detail_null when a value it shows is missing. Without a detail_null, an unknown row has no text, and the page shows it as no data.

A row reads label, issue, the three detail_ texts, action and actions, and nothing else. There is no warn state, and a row cannot skip itself: gate the whole step instead. The validator warns about any other key, because it does nothing.

#Warehouse

An enrichment is a fact Recued works out ahead of time and sticks onto a record. The enrichment-or-fetch transform reads one. It returns the stored fact when one exists, is fresh, and is trusted enough. If not, it tells your fallback step to fetch a fresh one.

It returns { value, source, confidence?, fallback? }. source is one of enrichment, miss, stale, low_trust, unparseable_ref, or no_runtime.

#Related

• Recipe schema: the steps these tools run inside.
• Authoring recipes: when to use a transform and when to use an AI step.