Something lands in memory
Trigger · Stable
Run when a document, email or note lands in memory — only for the ones that came from where you say and are the file types you say.
Fires once a document has landed in memory and its text and names have been read (the memory.item.annotated event), filtered before any run exists: a document that does not match never becomes a run. Two filters, both “is one of”, both optional — Came from (the channel: email, upload, connector, a flow, Scout, …, optionally narrowed to which mailbox, connector or flow) and File type (email, PDF, image, spreadsheet, …, read from the file itself). A flow never fires on a document it stored itself, nor on one written by an agent it starts (that agent’s own note would wake it again). Context keys: memory_item_id (the document id — read it with Get memory item or Memory fetch file), payload, and trigger.memory_item.{id, name, origin, file_type, mime_type, filename}.
When to use
Section titled “When to use”Use when a Job should read documents as they arrive in memory — invoices forwarded by email, photos a flow stores from a device, uploaded spreadsheets. Set the filters to what the Job acts on, so it runs on those documents and not on every item memory keeps; the Job decides the rest (its recipe claims the document, then reads it).
When not to use
Section titled “When not to use”If the flow should start from the email itself — its sender, subject and attachments as context — use Incoming email. For other platform events use Event; for a timed sweep over memory use Schedule (Cron). Heads-up: a run from this trigger has no signed-in user, so {{ user.* }} and {{ actor.* }} are empty. Bind any recipient or owner — a send_email to, a deal owner_actor_id, a task assigned_to — to a concrete email or actor_id, never {{ user.* }}; or collect it at run time with an Ask-a-person step.
Inputs
Section titled “Inputs”origin
Section titled “origin”list · optional · one of email_inbound, upload, paste, integration, flow, note, scout, webhook, mbox_import, unknown
The channel memory recorded when the document arrived — not what the document is. An email and its attachments came from Email; a file a flow’s Store document step saved came from Stored by a flow.
Example: email_inbound
- ✓
email_inbound— invoices and delivery notes your suppliers email in - ✓
flow— photos a device posts to a flow that stores them - ✗
invoice— that is a kind of document, not a channel — let the Job’s recipe decide what a document is
origin_detail
Section titled “origin_detail”list · optional
fnmatch patterns, case-insensitive, any one matching is enough. Email records every To and Cc address; a flow records its name and id; a connector records what it read from.
Example: invoices@*
- ✓
invoices@*— only mail sent to the invoices mailbox - ✓
Delivery ingest— only what the Delivery ingest flow stored - ✗
*@supplier.cz— that is a sender — this filter is about where the document arrived, not who sent it
file_type
Section titled “file_type”list · optional · one of email, pdf, image, spreadsheet, document, presentation, text, audio, other
Read from the file itself (its MIME type), never guessed from its content. Email is the email itself; its attachments are PDFs, images, spreadsheets and so on.
Example: pdf
- ✓
image— photos of delivery notes - ✓
email— a Job that answers emails - ✗
receipt— a kind of document, not a file type — the Job’s recipe decides that
Auto-generated from the skill registry (load_skills()). Do not edit by hand.