Operations reference

#Operations reference

A flow is a sequence of steps. Each step has a unique key, an operation type, and a configuration object:

{
	"key": "fetch_orders",
	"type": "crud_read",
	"config": {
		"collection": "orders",
		"filter": { "status": "pending" },
		"limit": 100
	}
}

Most configuration fields accept template expressions such as {{$trigger.body.id}} or {{= $sum($steps.fetch_orders.records.total) }} — see expressions. Each step's output is available to later steps as $steps.<step_key>.

the value you might expect. In particular run_script wraps the script's return value, so the path is {{$steps.<step_key>.output.<field>}} — not {{$steps.<step_key>.<field>}}. Each operation's output shape is listed with it below.

A template path that doesn't resolve produces an empty value rather than an error, and writes the reason to the step log — check there first if a field arrives unexpectedly blank.

#Step-level settings

Every step, regardless of operation type, supports these optional settings:

Setting	Values	Description
condition	rule tree	Skip the step when the condition evaluates to false
onError	stop, continue, retry, fallback	What to do when the step fails
retryConfig	{ maxAttempts, delayMs, backoff }	Retry settings (backoff: fixed or exponential) — used with onError: retry
fallbackValue	any	Value used as the step output when it fails — used with onError: fallback

#Flow options

Options apply to the whole flow:

Option	Default	Description
timeout_ms	30000	Maximum execution time for the entire flow
max_retries	0	Automatic retries if the flow fails
concurrency	sequential	sequential (one instance at a time), parallel, or a number of concurrent instances
error_handling	stop_on_first	stop_on_first or continue_all
log_level	full	full, errors_only, or none

#Data operations

#crud_action — Collection Records

Create, read, update, delete, or query records in a collection. One operation, five actions — the fields you configure depend on the action.

Field	Type	Required	Description
collection	string	Yes	Target collection
action	enum	Yes	create, read, update, delete, or query (default: read)
data	object	create/update	Field values for the record
id	string	No	Single record ID (update/delete)
ids	array	No	Multiple record IDs (update/delete)
filter	object	No	Filter criteria, e.g. {"status": "active"} (read/update/delete/query)
sort	array	No	Array of {field, order} objects (read/query)
limit	number	No	Maximum records to return, 1–10000 (default: 25)
offset	number	No	Records to skip (read)
fields	array	No	Specific fields to return; empty means all (read)
select	array	No	Fields to include in query results
groupBy	array	No	Fields to group query results by
aggregate	array	No	Array of {field, fn} where fn is count, sum, avg, min, or max
permanent	boolean	No	Permanently delete instead of soft-delete (default: false)

The five operations below are focused variants of crud_action. They behave identically — pick whichever reads better in your flow.

#crud_create — Create Record

Field	Type	Required	Description
collection	string	Yes	Target collection
data	object or array	Yes	Field values for the new record(s)

#crud_read — Read Records

Field	Type	Required	Description
collection	string	Yes	Target collection
filter	object	No	Filter criteria
sort	array	No	Array of {field, order} objects
limit	number	No	1–1000 (default: 25)
offset	number	No	Records to skip (default: 0)
fields	array	No	Specific fields to return

#crud_update — Update Records

Field	Type	Required	Description
collection	string	Yes	Target collection
data	object	Yes	Fields and values to update
id	string	No	Single record ID
ids	array	No	Multiple record IDs
filter	object	No	Filter matching the records to update

#crud_delete — Delete Records

Field	Type	Required	Description
collection	string	Yes	Target collection
id	string	No	Single record ID
ids	array	No	Multiple record IDs
filter	object	No	Filter matching the records to delete
permanent	boolean	No	Permanently delete instead of soft-delete (default: false)

#crud_query — Query Records

Aggregation and grouping over a collection.

Field	Type	Required	Description
collection	string	Yes	Target collection
select	array	No	Fields to include in results
filter	object	No	Filter criteria
groupBy	array	No	Fields to group by
aggregate	array	No	Array of {field, fn} — count, sum, avg, min, max
sort	array	No	Sort order
limit	number	No	1–10000 (default: 100)

#build_search_index — Build Search Index

Compiles a search index from its source collections into an R2 SQLite file.

Field	Type	Required	Description
indexName	string	Yes	The name of the search index to build

#Logic operations

#condition — Condition

If/else branching based on a rule tree. Rules compare a field against a value with operators such as eq, neq, gt, gte, lt, lte, in, nin, contains, starts_with, ends_with, is_null, is_empty, and matches, combined with _and/_or groups.

Field	Type	Required	Description
rules	object	Yes	Condition rule tree — use the visual builder or edit the JSON directly

#switch — Switch

Multi-branch routing based on matching a value against cases.

Field	Type	Required	Description
value	template	Yes	Value to match, e.g. {{$trigger.body.status}}
cases	array	Yes	Array of {match, steps} objects
default	array	No	Steps to run if no case matches

#loop — Loop

Iterates over an array, executing nested operations for each item. Inside the loop, reference the current item as {{$steps.<loop_key>.item}} and the zero-based index as {{$steps.<loop_key>.index}}.

Field	Type	Required	Description
items	template	Yes	Array to iterate over, e.g. {{$steps.fetch_orders.records}}
mode	enum	No	serial (default), parallel, or batch
batchSize	number	No	Items per batch in batch mode, 1–50 (default: 10)
operations	array	Yes	Operations to execute for each item

#stop — Stop Flow

Halts the flow with a success or error status. In an event_filter flow, stopping with an error status rejects the triggering write.

Field	Type	Required	Description
status	enum	No	success (default) or error
message	template	No	Exit message (default: "Flow execution stopped.")

#throw_error — Throw Error

Aborts the flow with an error message and HTTP status code.

Field	Type	Required	Description
message	template	Yes	Error message
statusCode	number	No	HTTP status code, 100–599 (default: 400)

#set_variable — Set Variable

Stores a value for later steps to read via {{$steps.<step_key>}}.

Field	Type	Required	Description
key	string	Yes	Variable name
value	template	Yes	Value to store

#trigger_flow — Trigger Flow

Executes another flow. Sub-flows nest up to 8 levels deep; circular calls are rejected.

Field	Type	Required	Description
flowId	string	Yes	The flow to trigger
data	object	No	Payload passed to the triggered flow as $trigger.body
async	boolean	No	Fire and forget — don't wait for the sub-flow (default: false)

#Transform and utility operations

#transform — Transform Data

Reshapes data through a pipeline of operations.

Field	Type	Required	Description
input	template	Yes	Input data, e.g. {{$steps.fetch_orders.records}}
operations	array	Yes	Pipeline of operations: map, filter, pick, omit, merge, flatten, unique, sort, reverse, slice, groupBy

#json_parse — JSON Parse

Parses a JSON string and optionally extracts a value by path.

Field	Type	Required	Description
input	template	Yes	String or object to parse
path	string	No	Dot-path to extract from the parsed result, e.g. data.items

#text_format — Text Format

String manipulation via a template, a list of text operations, or both.

Field	Type	Required	Description
template	text	No	Text template with variable substitution
operations	array	No	Text operations: concat, split, replace, regex_replace, trim, uppercase, lowercase, substring, pad

#math — Math

Mathematical operations over an array of values or a formula.

Field	Type	Required	Description
operation	enum	Yes	sum, avg, min, max, round, floor, ceil, abs, or formula
values	array	No	Numbers to operate on
formula	template	No	Formula expression, e.g. (a + b) * c (formula operation)
precision	number	No	Decimal precision, 0–20

#datetime — Date / Time

Date and time calculations.

Field	Type	Required	Description
operation	enum	Yes	now, format, parse, add, subtract, diff, startOf, endOf
value	template	No	Input date/time string
amount	number	No	Units to add or subtract
unit	enum	No	milliseconds, seconds, minutes, hours, days, weeks, months, years
format	string	No	Format pattern, e.g. YYYY-MM-DD HH:mm:ss
timezone	string	No	Timezone name, e.g. Australia/Sydney

#log — Log

Writes a message to the flow run log — useful for debugging.

Field	Type	Required	Description
message	template	Yes	Message to log
level	enum	No	info (default), warn, error, debug

#delay — Delay

Pauses the flow.

Field	Type	Required	Description
ms	number	Yes	Milliseconds to wait, 0–25000 (default: 1000)

#render_lens_template — Render Lens Template

Renders a Lens template to HTML, CSV, or Markdown, optionally saving the result to R2.

Field	Type	Required	Description
templateId	string	Yes	ID of the Lens template
format	enum	No	html (default), csv, or md
urlParams	object	No	Query parameters passed to the renderer
saveToR2	boolean	No	Save the output to R2 (default: false)
r2Key	string	No	Custom R2 key for the output file

#AI operations

AI operations count against your project's daily AI token quota. When the quota is exhausted, AI steps fail until the next day.

Requirements: AI operations run on Workers AI via the gateway's AI binding, which the standard deploy templates include. If a step fails with "AI provider … does not support" or the binding is absent (check GET /health), your instance's worker config is missing the [ai] block. provider currently supports cloudflare only — other providers are reserved for future use.

ai_classify is production-ready and a common building block for content moderation: classify a post's text into labels like ok / spam / abuse in a flow triggered on record create, then act on the result (flag, soft-delete, or queue for review) in subsequent steps.

#ai_generate — AI Generate Text

Generates text, with optional system prompt and JSON mode.

Field	Type	Required	Description
userPrompt	text	Yes	The prompt to send to the model
systemPrompt	text	No	System/instruction prompt
model	string	No	Model ID, e.g. @cf/meta/llama-3.1-8b-instruct
maxTokens	number	No	Maximum tokens to generate
temperature	number	No	0–2 (default: 0.7)
responseFormat	enum	No	text (default) or json
jsonSchema	object	No	Expected JSON response structure (JSON mode)
provider	string	No	AI provider (default: cloudflare)

#ai_classify — AI Classify Text

Classifies text into one of several labels. Returns the best label, a confidence score, and scores for every label.

Field	Type	Required	Description
text	template	Yes	Text to classify
labels	array	Yes	At least 2 classification labels
model	string	No	Model ID
provider	string	No	AI provider

#ai_extract — AI Extract Structured Data

Extracts structured data from text according to a schema.

Field	Type	Required	Description
text	template	Yes	Source text
schema	object	Yes	Object describing what to extract
instructions	text	No	Extra extraction instructions
model	string	No	Model ID
provider	string	No	AI provider

#ai_image — AI Generate Image

Generates an image from a text prompt.

Field	Type	Required	Description
prompt	text	Yes	Image prompt
model	string	No	Model ID
width	number	No	Image width in pixels
height	number	No	Image height in pixels

#ai_review — AI Review

Reviews text against criteria and returns a score, feedback, issues, and suggestions.

Field	Type	Required	Description
content	text	Yes	Content to review
criteria	array	No	Criteria to evaluate against
threshold	number	No	Score threshold for auto-approval, 0–1 (default: 0.7)
model	string	No	Model ID

#ai_summarize — AI Summarize

Summarises text with a configurable length and style.

Field	Type	Required	Description
text	text	Yes	Text to summarise
style	enum	No	brief (default), detailed, or bullets
maxLength	number	No	Maximum character length of the summary
model	string	No	Model ID

#ai_translate — AI Translate

Translates text to a target language.

Field	Type	Required	Description
text	text	Yes	Text to translate
targetLanguage	string	Yes	Target language, e.g. Spanish
sourceLanguage	string	No	Source language (auto-detected if empty)
model	string	No	Model ID

#File operations

#file_read — Read File

Reads content from your project's R2 object storage.

Field	Type	Required	Description
key	template	Yes	File key, e.g. uploads/report.csv
encoding	enum	No	text (default), base64, or json

#file_upload — Upload File

Uploads content to R2 object storage.

Field	Type	Required	Description
key	template	Yes	File key, e.g. exports/report.csv
content	text	Yes	File content (text or base64-encoded)
contentType	string	No	MIME type (default: application/octet-stream)
metadata	object	No	Custom metadata key/value pairs

#file_transform — Transform File

Applies image transformations to stored files, using the server-side resizer. The source is read from wherever the file actually lives — a container-scoped or tenant-dedicated bucket included.

Field	Type	Required	Description
sourceKey	template	Yes	Source file key
destKey	template	No	Destination key (derived from the source if empty)
operations	array	Yes	Array of {type, ...options}: resize, format, quality

What the resizer can do:

• resize — width, height (either or both) and fit. contain (default) and scale-down fit within the box and preserve aspect ratio. cover and crop need a crop the resizer cannot perform.
• format — png, jpeg or webp. With no format operation the source's own format is kept; a source in a format that can be read but not written (gif, bmp, tiff) must name one. avif cannot be written.
• quality — webp output only.

A transform the resizer cannot perform fails the step, and nothing is written to the destination key. It never falls back to copying the source, which is what it used to do: the output was a byte-identical copy at a new key and the run reported success.

#s3_action — S3 / R2 File

Put, get, delete, list, or check files in any S3-compatible bucket via a storage connection.

Field	Type	Required	Description
connectionId	string	Yes	Storage connection (create them in Settings → Storage Connections)
action	enum	Yes	put, get, delete, list, or head
key	template	put/get/delete/head	Object key
content	text	put	Content to upload (text or base64)
contentType	string	No	MIME type for uploads
encoding	enum	No	For get: text, base64, or json
prefix	string	No	Key prefix for list
maxKeys	number	No	Maximum objects to list

#Network operations

Outbound requests are blocked from reaching internal and private network addresses.

#http_request — HTTP Request

Makes an outbound HTTP request. The output contains status, statusText, headers, body, and durationMs.

Field	Type	Required	Description
url	template	Yes	Target URL (http or https)
method	enum	No	GET (default), POST, PUT, PATCH, DELETE, HEAD, OPTIONS
headers	object	No	Request headers
body	any	No	Request body — objects are JSON-encoded automatically
responseType	enum	No	json (default), text, or binary (base64)
timeout_ms	number	No	100–30000 (default: 10000)
auth	object	No	{"type":"bearer","token":"…"}, {"type":"basic","username":"…","password":"…"}, or {"type":"api_key","header":"X-API-Key","token":"…"}

#graphql_request — GraphQL Request

Sends a GraphQL query or mutation and returns the data and errors.

Field	Type	Required	Description
url	template	Yes	GraphQL endpoint URL
query	code	Yes	Query or mutation string
variables	object	No	Query variables
headers	object	No	Request headers
operationName	string	No	For multi-operation documents
timeout_ms	number	No	100–30000 (default: 10000)

#webhook_dispatch — Webhook Dispatch

POSTs a JSON payload to a webhook URL with optional HMAC-SHA256 signing. See sending webhooks.

Field	Type	Required	Description
url	template	Yes	Destination webhook URL
payload	object	No	JSON payload (defaults to the trigger body)
headers	object	No	Extra request headers
secret	string	No	HMAC-SHA256 shared secret — adds X-EmuView-Signature and X-EmuView-Timestamp headers
timeout_ms	number	No	100–30000 (default: 10000)

#Script operations

#run_expression — Run Expression

Evaluates a JSONata expression. Always available — no extra setup required.

Field	Type	Required	Description
expression	code	Yes	JSONata expression. Access data via $trigger, $steps, $user, $data
input	object	No	Additional data merged into $data
timeout_ms	number	No	100–10000 (default: 5000)

#run_script — Run Script (JavaScript)

Executes custom JavaScript in an isolated sandbox. See scripts for the sandbox environment, limits, and requirements.

Field	Type	Required	Description
code	code	No*	Inline JavaScript
scriptId	string	No*	Saved script to execute
input	object	No	Data merged into $data inside the script
timeout_ms	number	No	100–30000, capped at the project script CPU quota
allowed_domains	array	No	Domains the script may fetch (empty = all blocked)

* Provide either code or scriptId.

#Communication operations

#send_email — Send Email

Sends a transactional email. Emails count against the project's daily email quota.

Field	Type	Required	Description
to	template	Yes	Single address or array of addresses
subject	template	Yes	Subject line
html	text	No	HTML body
text	text	No	Plain text body
from	string	No	From address
replyTo	string	No	Reply-To address
template	text	No	Template with {{key}} placeholders
templateData	object	No	Values for the template placeholders

Recipients are not restricted by the operation — any valid address or array of addresses works, gated only by the daily email quota (each recipient counts once). Templates support dot-path keys ({{user.name}}); a rendered template containing HTML tags is sent as the HTML body, otherwise as text.

Output

Field	Type	Description
messageId	string	Provider message id where available, otherwise a generated id
status	string	sent
to	string | string[]	Echo of the recipients
channel	string	email_sender_binding | email_binding | rest_api — which channel delivered
quotaRemaining	number	Daily allowance left after this send

Branch on channel to detect the active transport from inside a flow, without calling /health.

Errors

Every failure is typed and carries a stable code. Quota exhaustion is always a refusal, never a silent drop — the operation raises before touching a transport, so nothing is delivered and a multi-recipient batch is never partially sent.

code	Meaning	Retryable
email_quota_exceeded	This project's emails_per_day would be exceeded. Carries limit, used, requested, remaining.	No — resets daily
email_upstream_quota_exceeded	The provider rate-limited or hit its own daily cap (e.g. HTTP 429).	Yes, with backoff
email_recipient_not_allowed	Recipient unreachable on this channel — typically an unverified destination on a channel whose sending domain is not onboarded.	No
email_no_channel	Nothing is configured to send with. Message names exactly what to configure.	No
email_no_capacity	Transports exist, but every one is out of allowance, outside its sending hours, or in a cool-off after a failure.	Yes
email_send_failed	Delivery failed for another reason.	Only if 5xx

email_quota_exceeded fires against the EmuView per-project quota; email_upstream_quota_exceeded fires against the provider's. Distinguish them before retrying — the first cannot succeed again today, the second usually can.

#send_notification — Send Notification

Sends an in-app notification to a user or the project log.

Field	Type	Required	Description
title	template	Yes	Notification title
message	template	Yes	Notification body
type	enum	No	info (default), success, warning, error
userId	template	No	Target user (defaults to the flow runner)
link	template	No	Link opened when the notification is clicked

#System operations

These operations only run inside system-managed flows. They appear here for completeness — you can't add them to your own flows.

Operation	Purpose
system_archival	Archives Cloudflare GraphQL metrics to R2 for historical analysis
system_limit_check	Checks current usage against plan thresholds and creates alerts
system_monthly_summary	Aggregates daily R2 archives into a monthly summary