Workflows
Automate tasks, invoices, and messages with a graph.
Open Workflows in the sidebar. This is the in-app editor. Keys and outbound webhooks are in the developer hub. Field paths and {{ }} live in Expressions.
You need Manage workflows to create, edit, and run these. That is a workspace permission — see roles. Workflows run as the workspace (owner access), not as you. A member who can edit a graph can still create tasks and invoices the owner could create.
Create one
Click New workflow, name it, and Create. It starts as Draft with an empty canvas. Click Add Trigger in the middle of the editor to choose how it starts.
| Status | What it does |
|---|---|
| Draft | Edit freely. Test run works. Live events and inbound webhooks do not fire. |
| Active | Live events, schedule, and inbound webhook fire. Save (⌘S) before you Activate. |
| Paused | Stays saved. Nothing live runs until you resume. |
| Archived | Hidden from the working list. |
Test run vs live
Test run in the editor executes the graph for real — it can create tasks, send HTTP, and post in a channel. It uses the trigger’s test payload JSON as input. Draft workflows can be test-run.
Live runs (platform events, schedule, inbound webhook) only happen when the workflow is Active. Edit the test payload so it looks like a live event before you activate.
Build the graph
Drag nodes from the palette, connect them, and click a node to edit it in the inspector. The Input panel lists fields you can drag into expressions and templates.
Triggers — Manual, inbound webhook, schedule (cron + timezone), or a platform event (project, task, customer, contact, sequence, event, form, funnel, pipeline, opportunity, invoice, payment, expense, channel, message, Instagram, email, community, subscription). Every webhook event can start a workflow.
Flow — If, Switch, Match, Conditions, Foreach, Wait. How to write the expressions: Expressions.
Actions — Log, HTTP request, Send Instagram message, Private reply to comment, Reply to Instagram comment, create/update/get for tasks, customers, contacts, forms, pipelines, opportunities, invoices, and payments, Record purchase, Reverse purchase, Get submission, Finalize invoice, Accept waitlist entry, Reject waitlist entry, and Channel message. There is no “mark invoice paid” node, and no project actions. Record purchase is idempotent on externalRef and is how any signed outside checkout becomes a sale (webhook trigger → record purchase). Amounts are integer minor units; "29.00" is also accepted.
AI — AI Agent plus a connected OpenAI or Anthropic provider (required), and optional tools (HTTP request, Set variable, Run workflow, MCP, and Planeward actions such as Create task). Attach an OpenAI or Anthropic secret on the provider node.
Inbound webhook
Add an Inbound webhook trigger. The inspector shows the public URL with a Copy button:
POST https://app.planeward.com/api/webhooks/workflows/<workspaceId>/<path>The path is a UUID, generated for you and read-only. Each path is unique in the workspace. A pasted or duplicated trigger gets a new path, and a trigger saved without one gets one on save. Regenerate issues a new path. Once you save, requests to the old one return 404, so update every caller first. Only an Active workflow accepts the request.
The URL now includes your workspace ID
The old /api/webhooks/workflows/<path> URL keeps working until
November 28, 2026. Its responses carry Deprecation, Sunset, and a
Link header that points at the new URL. From that date it returns 410 Gone. Copy the new URL from the inspector and update every caller before
then.
Send JSON with Content-Type: application/json. Anything else is stored as an empty body.
POST /api/webhooks/workflows/<workspaceId>/<path> HTTP/1.1
Content-Type: application/json
Idempotency-Key: order-123
{"email":"[email protected]","name":"Ada"}A matching run returns 202:
{ "runId": "…", "status": "accepted" }When the trigger's filter rejects the request, it still returns 202, but no run is created:
{ "filtered": true }Unknown path, a path from another workspace, or a draft/paused workflow is 404. Failed header or signature auth is 401. Invalid JSON is 400. A body larger than 1 MB is 413.
Auth mode on the trigger:
- None — anyone who has the URL can start a run
- Header — pick a secret and a header name (default
x-workflow-secret). The request must send that exact value. - HMAC signature — pick a secret and a header name (default
x-signature). The sender signs the raw request body with HMAC-SHA256 using that secret and sends the digest ashex(default) orbase64. Asha256=prefix is accepted. Use this when the sender (Stripe-style, Shopify-style, or your own service) signs payloads instead of sending a shared header.
Optional Idempotency-Key header: the same key will not start a second run.
In the graph, the payload is:
input.body.email;
input.body.items;
input.headers["content-type"];To record a sale from an outside checkout, add Record purchase after the webhook. Map externalRef to the sender's payment id so retries return already_recorded. Map amount (minor units or "29.00"), email/phone/name, and optional sourceLabel. Turn Send purchases to Meta on the site if you want a live Purchase event.
Platform events
Pick a Task created, Invoice overdue, and so on from the palette (or a generic Platform event and choose the event). Live envelope:
input.event; // "task.created"
input.data;
input.timestamp;input.data is small. Use Get task / Get invoice / Get customer / Get contact / Get form / Get submission / Get pipeline / Get opportunity / Get payment when you need the rest of the record. Community membership, receipt, and subscription events include customerId; use Get customer / Get payment for the rest. There is no Get-community action.
Every event's data has a fixed schema. The Input panel shows a sample, and planeward-workflow-node-types returns the JSON Schema for each event.
Updated events. Every *.updated trigger (task.updated, project.updated, contact.updated, and so on) sends { id, <parent ids>, old, new, changes }. old and new hold only the fields that changed, and changes lists their names. Read input.data.new.statusId, not input.data.statusId. Field lists per event: Webhooks. Examples: Expressions.
| Event | input.data |
|---|---|
any *.updated | id, parent ids, old, new, changes |
project.created / deleted | id, name |
task.created | id, projectId, parentId, title, statusId, priority, startDate, dueDate, completedAt |
task.deleted | id, projectId, parentId, title |
customer.created | id, name |
customer.deleted | id |
contact.created | id, name |
contact.deleted | id |
contact.unsubscribed | id, contactId, deliveryId, source (unsubscribe_link, manual, complaint), timestamp |
contact.dm_opt_out / marketing_consent | id, contactId, timestamp |
contact.merged | id (the kept contact), keepContactId, mergedContactId, timestamp |
sequence.enrolled / stopped / paused / resumed / completed | id, sequenceId, sequenceKey, lane, contactId, opportunityId, status, stopReason, currentStep (and previousSequenceKey on enrolled) |
event.registered / attendance_classified | id, eventId, contactId, opportunityId, attendance |
form.submitted | id, formId, contactId, customerId, opportunityId, source |
funnel.started / step_viewed / completed | funnelId, siteId, stepIndex, pageId |
pipeline.created | id, name |
pipeline.deleted | id |
opportunity.created | id, name |
opportunity.stage_changed | id, name, pipelineId, stageId, stageName, previousStageId, stageType, contactId, customerId, ownerMembershipId, value |
opportunity.deleted | id |
invoice.created | id, invoiceNumber |
invoice.deleted / finalized | id, invoiceNumber |
invoice.paid | id |
invoice.due / overdue | id, invoiceNumber, customerId, dueAt |
payment.created | id, amount, status |
payment.completed | id, amount, currency, source, sourceLabel, contactId, customerId, opportunityId, invoiceId, externalRef |
payment.refunded | id, amount, currency, contactId, opportunityId, externalRef, reason |
expense.created / deleted | id, amount, currency, spentAt, description, vendor, categoryId, adjustsExpenseId, version |
expense.adjusted | Same as above plus reason |
expense.budget_exceeded | categoryId, categoryName, month, budget, spent |
channel.created | id, name, slug |
message.created | id, channelId, threadParentId, authorMembershipId, authorName, body, createdAt |
message.deleted | id, channelId |
instagram.message.received | id, contactId, instagramConnectionId, instagramUsername, senderId, text, source, postbackPayload, storyId, storyUrl, referralRef, timestamp |
instagram.comment.created | id, contactId, instagramConnectionId, instagramUsername, commenterId, commenterUsername, text, mediaId, parentCommentId |
email.delivery.* | deliveryId, emailType, campaignId, contactId, toEmail, timestamp |
email.campaign.completed | campaignId, timestamp |
community.membership.* | id, communityId, customerId, source, status |
community.receipt.submitted / approved / rejected | id, communityId, customerId, productPlanId, status |
subscription.renewal_upcoming | id, customerId, customerName, customerEmail, productId, productName, productPlanId, planName, price, currency, interval, intervalCount, communityId, periodEnd, hoursBefore, cancelAtPeriodEnd, managedBy |
subscription.started / renewed | id, customerId, customerName, customerEmail, productId, productName, productPlanId, planName, price, currency, interval, intervalCount, communityId, status, previousStatus, periodEnd, cancelAtPeriodEnd, managedBy, complimentary, paidCycles, maxCycles |
subscription.completed | Same as above, plus lifetimeAfterFinalCycle |
subscription.cancel_scheduled / canceled / expired / terminated | Same as above, without paidCycles and maxCycles |
community.course.published | id, communityId, title |
community.waitlist.joined | id, communityId, productId, email, contactId, status |
community.waitlist.accepted | id, communityId, productId, email, contactId, customerId, status |
community.waitlist.rejected | id, communityId, productId, email, contactId, status |
invoice.due and invoice.overdue fire on a daily job (UTC) for finalized, unpaid invoices. subscription.* events fire for every plan (Stripe, Xendit, or manual); communityId is set only for plans of products that grant community access. Receipt communityId is null when the plan is not linked, and waitlist communityId is null for a workspace with no community. The receipt and waitlist events come from Commerce and keep their community.* names. event.registered also fires when a payment for a product linked to the event registers the buyer, so a workflow on it can send a ticket confirmation (Paid events).
To choose which events start the workflow, add trigger filters. To branch on the event inside the graph, use If:
input.data.changes.includes("statusId") &&
input.data.new.statusId === "status_done";
outputs["Get invoice"].status === "overdue";Skipped runs. Trigger filters run first; an event they reject never reaches the Match gate. When an event trigger is followed only by Match nodes and the event lands on a Match branch with nothing connected (usually an unconnected Default), no run is created. It does not show in Executions and does not count toward run limits. This keeps Instagram keyword flows from filling run history with every non-matching DM or comment. To still record those events, connect something to Default. If the Match expression errors, the run is created anyway so you can see the error.
Subscription renewal upcoming
Subscription renewal upcoming (subscription.renewal_upcoming) fires once per period for every active subscription that has a period end, community plan or not, a set time before the period ends.
Set Send before period ends on the trigger in hours or days, from 1 hour up to 90 days (for example 24 hours or 3 days). For several reminders, add one trigger per lead time, or use separate workflows.
- Timing follows each subscription's own end time (within about a minute), not a daily job.
- A renewal starts a new period, which gets its own reminder.
- Reminders whose time had already passed when you activated the workflow are not sent.
- It also fires for subscriptions set to cancel at period end. Branch on
input.data.cancelAtPeriodEndwith If or Match to send "renews soon" or "expires soon" copy.
price is in minor units (2900 is $29.00). customerEmail and communityId can be null. hoursBefore is the trigger's lead time in hours. managedBy is MANUAL, STRIPE, or XENDIT.
input.data.cancelAtPeriodEnd; // true → "expires soon"
input.data.periodEnd; // "2026-12-01T00:00:00.000Z"Subscription lifecycle
These events fire for every plan, whether Stripe, Xendit, or manual, with or without a community. Filter by productId to target one product, or by communityId to target community plans.
| Event | Fires when |
|---|---|
Subscription started (subscription.started) | The first paid activation, a rejoin after the previous subscription ended (its status was not ACTIVE), or a staff grant (complimentary is true) |
Subscription renewed (subscription.renewed) | A continuing paid cycle on a subscription that was already ACTIVE |
Subscription fully paid (subscription.completed) | A payment reached the plan's cycle cap. Fires alongside that payment's started or renewed event, and may arrive before it |
Subscription cancellation scheduled (subscription.cancel_scheduled) | The member or the processor set the subscription to cancel at period end. It stays ACTIVE until periodEnd |
Subscription canceled (subscription.canceled) | The processor stopped billing: the Stripe subscription was deleted, or the Xendit plan was inactivated. Staff revoking a grant fires it too. Not sent for a subscription that already expired |
Subscription expired (subscription.expired) | The paid period ran out (see below), or a Xendit cycle failed after the period ended |
Subscription terminated (subscription.terminated) | A refund or chargeback ended the subscription immediately |
- Expiry is checked hourly. Stripe and Xendit plans that are still renewing expire 72 hours after
periodEnd, which gives the processor time to collect the renewal. Manual plans and plans set to cancel expire atperiodEnd. statusis the status after the change andpreviousStatusthe one before it (can be null).startedandrenewedalso sendpaidCyclesandmaxCycles.completedsends both pluslifetimeAfterFinalCycle.complimentaryistruefor a staff grant: free, lifetime access with no payment. A grant firessubscription.startedwithmanagedByMANUAL,paidCycles0, and noperiodEnd. To run only on paid starts, add a field condition oncomplimentaryEqualsfalse. A later payment for the same plan never touches the grant: it starts or continues a paid subscription of its own, with its own events.communityIdis set for plans of products that grant community access and null otherwise. To run only for community plans, add a field condition oncommunityIdwith Not empty.priceis in minor units,customerEmail,communityId, andperiodEndcan be null, andmanagedByisMANUAL,STRIPE, orXENDIT, as in renewal upcoming.
input.data.productId; // filter on this, or branch with Match
input.data.previousStatus; // "EXPIRED" on a rejoin
input.data.periodEnd; // when access ends for cancel_scheduledMigrating from community.subscription.* (2026-11-05)
The community.subscription.* triggers are removed. Workflows that used one
were switched to the matching subscription.* event, with a field condition
communityId Not empty so they still run only for plans of products that grant community access.
input.data is now the full lifecycle payload above.
| Removed | Replacement |
|---|---|
community.subscription.started | subscription.started |
community.subscription.renewed | subscription.renewed |
community.subscription.expired | subscription.expired |
community.subscription.completed | subscription.completed |
A rejoin (the subscription ended, then was paid again) now fires subscription.started, not renewed. Workflows that were on community.subscription.renewed no longer run for rejoins; add a subscription.started trigger if they should.
Trigger filters
Event and inbound webhook triggers have a Filter section. A filter decides whether an event starts the workflow at all. When it rejects an event, no run is created: nothing shows in Executions and nothing counts toward run limits. A webhook that is filtered out still replies 202, with { "filtered": true } instead of a runId.
A filter is one list of conditions. Set the toggle to All (every condition must pass) or Any (one is enough). A trigger with no conditions starts on every event. A filter holds up to 20 conditions, and each condition up to 100 values or ids.
A filter is not authentication. On a webhook with no auth, anyone who has the URL can tell whether a request was filtered out from the { "filtered": true } reply. Use header or HMAC auth for anything sensitive.
Field conditions
A field condition compares one value in the payload.
- Path — for an event, a path inside
input.data, such asproductIdornew.statusId. For a webhook, start withbodyorheaders, such asbody.typeorheaders.x-source(header names ignore case). Under the conditions, Browse payload fields lists the fields in the trigger's test payload: drag a field onto a condition, or click it to use it. You can also type or paste a path. - Operator — Equals, Not equal, Contains, Any of, None of, Empty, Not empty, Greater than, Greater than or equal, Less than, Less than or equal.
- Value — one value, a list for Any of and None of, or nothing for Empty and Not empty. Paths that hold an id (product, plan, pipeline, stage, sequence, contact, customer) get a picker instead of a text box.
Text matching in field conditions ignores case. The number operators compare the field as a number, and a value that is not a number never matches ("abc" is neither greater nor less than 5). The Match node supports the same operators, so it can branch on numbers too.
Contact and customer conditions
These check the person behind the event, live in your CRM. Each one can be flipped between has and lacks:
- Tag — the contact, or the customer, has or lacks any of the chosen tags. Tags match exactly, including case (
VIPis notvip) - Subscription — the customer has or lacks an active subscription to any of the chosen products
- Deal — the contact has or lacks a deal in any of the chosen pipelines, optionally only in certain stages
- Sequence — the contact is or isn't actively enrolled in any of the chosen sequences
They are only offered for events that identify a contact or customer, such as form.submitted, payment.completed, subscription.started, or opportunity.stage_changed. Events like pipeline.created or expense.created offer field conditions only, and webhook triggers never offer them.
- Conditions are checked when the workflow is about to start, so they see the CRM as it is then, not as it was at the moment of the event.
- When the event's contact or customer can't be found, every has condition fails and every lacks condition passes.
Example
Start a welcome flow only when someone subscribes to your Coaching product and isn't already a VIP:
- Add an event trigger for Subscription started (
subscription.started). - Set the filter to All.
- Add a field condition:
productIdEquals Coaching. - Add a tag condition: customer lacks tag
vip.
Through the API (planeward-workflow-graph-update), the same filter is config.filter on the trigger:
{
"eventId": "subscription.started",
"filter": {
"match": "all",
"rules": [
{
"id": "r1",
"kind": "field",
"path": "productId",
"operator": "equals",
"values": ["<productId>"]
},
{
"id": "r2",
"kind": "lookup",
"lookup": "tag",
"subject": "customer",
"negate": true,
"ids": ["vip"]
}
]
}
}Operators in the API are equals, notEquals, contains, anyOf, noneOf, empty, notEmpty, gt, gte, lt, and lte. Lookups are tag (with subject contact or customer, ids are tag names), subscription (ids are product ids), deal (ids are pipeline ids, optional stageIds), and sequence (ids are sequence ids). negate: true means lacks. The old config.filters object (projectId, statusId, priority, changedFields, hoursBefore) is rejected with an error telling you to use config.filter and config.hoursBefore instead. Existing workflows were converted automatically.
Schedule
Cron plus a timezone (IANA name, default UTC). Example: 0 9 * * * in America/New_York is 9:00 every day in that zone.
A live run sends only the fire time:
input.scheduledAt; // "2026-08-29T09:00:00.000Z"The test payload may also include cron. Use Schedule for “every morning.” Use Wait for “a day after this event.”
Optional Ends at stops the schedule after that date and time. Set it on yearly or one-off crons (a launch reminder, a renewal date) so they do not fire again next year.
Wait
Set a number of seconds (1 second to 7 days in the inspector).
On an Active workflow, a Wait longer than 5 seconds parks the run and continues it when the time is up. A 3-day Wait really waits 3 days. The run shows Running in Executions until then, with the Wait's resumeAt in its output. Earlier node outputs stay available after the Wait. If the workflow is paused or archived during the Wait, the run stops as Cancelled instead of continuing.
A Wait is not durable in three cases, and sleeps at most 5 seconds before going on:
- the workflow is a Draft (so test runs stay quick)
- the Wait sits inside a Foreach or an AI tool
- the run is a dry run
Re-check the record after a long Wait (for example Get contact, then If the stop tag is gone) because things change while the run is parked. A run can contain up to 50 durable Waits. To chain more than 7 days, connect several Waits.
Actions
Most text fields are templates ({{ input.data.title }}). Search pickers can stay on a fixed project or channel; switch the field to Expression when the id comes from the run.
Create task — projectId, statusId, and title are required. Optional description, priority, assignees (JSON array of membership ids), Assignee role IDs (JSON) (assigneeRoleIdsJson, a JSON array of role ids), start date, due date. Only roles that can be assigned in the project are accepted.
Update / get task — taskId (usually {{ input.data.id }}). Update can change title, status, priority, dates, assignees, and assignee roles. Leave Assignee role IDs (JSON) empty to keep the current roles, or send [] to remove them all.
Create customer — name required. Optional email, phone, company, and tags (a JSON array or comma-separated list). Tags are case-sensitive (VIP and vip are different).
Update customer — customerId. Optional name, email, phone, company, tags, lifecycleStatus, and nextFollowUpAt (ISO date). A blank field leaves the current value in place, including address and role. A JSON tags array replaces the tags.
Add / set / remove customer tags — customerId and tags. Add keeps existing tags. Set replaces them, and an empty list clears them. Remove drops exact matches.
List customers — optional tag (one exact tag), search, and limit (1–100, default 50). Returns list and count.
Create / update / get contact — name required on create. Update and get take contactId. Optional email, phone, role, customerId, and tags (same rules as customers).
Add / set / remove contact tags — contactId and tags. Same add, replace, and remove behavior as customer tags.
List contacts — optional customerId, tag, search, and limit. Returns list and count.
Create / update / get form — name required on create. Update and get take formId. Optional description, slug, public slug, and status (DRAFT or PUBLISHED). fieldsJson / settingsJson replace the field tree or CRM/embed settings when set.
Get submission — submissionId (usually {{ input.data.id }} after Form submitted). Returns answers, source, and linked contact / customer / opportunity ids.
Create / update / get pipeline — name on create and update. Update and get take pipelineId. Create seeds the default stages.
Create opportunity — pipelineId and contactId required. Optional name, stageId (defaults to the first Open stage), customerId, value, ownerMembershipId, and nextFollowUpAt. Update opportunity takes opportunityId and can change name, stageId, contact, customer, value, owner, and follow-up date. Use stageId to move the card.
List opportunities — optional pipelineId, contactId, customerId, stageType (OPEN, WON, LOST, or a comma-separated list), search, and limit. Returns list and count. Use it to check for an open deal before creating a second one.
Create invoice — customerId and title. The node opens a draft with no line items. Finalize invoice takes invoiceId and locks it (same as Finalize in CRM). There is no mark-paid action.
Accept waitlist entry — entryId (usually {{ input.data.id }} after Community waitlist joined). Same as Accept on the product waitlist: unlocks the product's waitlist plans for checkout with that email, emits community.waitlist.accepted, and emails a sign-in link or code (the workspace's sign-in method). It gives no community access; the person still has to check out. Leave entryId blank after Waitlist joined to use that entry. Already-accepted entries succeed without changing anything.
Reject waitlist entry — entryId, filled in the same way as Accept. Same as Reject on the product waitlist: the applicant stays blocked from checkout and community.waitlist.rejected is emitted. To accept or reject applicants automatically, start from Community waitlist joined and branch on the application's answers before an Accept or Reject node.
Get invoice returns the record plus computed status: draft, pending, overdue, or paid. amount is integer cents (50000 is $500).
Record payment — customerId, amount (dollars with a decimal, or integer cents), optional invoice and method.
Log expense — amount (dollars with a decimal, or integer cents) and description. Optional categoryId, vendor, notes, and spentAt (YYYY-MM-DD, defaults to today). Update expense takes expenseId and changes only the fields you set. It fails for expenses in a closed month; correct those with an adjustment in the app. Get expense returns the record plus isLocked and adjustedTotal.
expense.budget_exceeded fires when a change pushes a category's spend for a month from at-or-under its monthly budget to over it.
Channel message — channelId and body.
Notify member (community.member.notify, in the Community group) — sends one member a bell and push notification titled with the community's name. It never sends email. The owner of the workflow needs Manage community.
memberId— a community membership id or a customer id. Leave it blank to use the trigger'smemberId, then itscustomerId. This works only when the workflow starts from a platform event; a webhook, manual or scheduled run never supplies the member, so set it there.body— required, plain text, 1–280 characters.link— optional. A path on the community site, such as/courses/abc. It can't be an external URL or start with//, and can't contain.or..segments or#. With no link, the notification opens the community home.
A member who isn't found, whose access is suspended or revoked, or whose plan has lapsed, is skipped and the step still succeeds, with skipped: true, a reason (not_a_member or member_not_active) and a message. A sent step returns memberId and notified: true. A retried step never notifies twice. Members control these in Settings › Notifications under Community updates (a push switch), and the workflow is not told if they turn it off. A dry run validates the fields without sending. Missing or invalid fields fail the step with a message that names the field.
Send Instagram message — reply in a DM thread. body is required (up to 1,000 characters). recipientId is the sender's Instagram user id; leave it blank after Instagram message received to reply to that sender. Instagram only allows replies within 24 hours of the person's last message. Set messageTag to HUMAN_AGENT for a reply a person wrote, sent up to 7 days after their last message; the Meta app needs Human Agent permission. Add a buttonLabel (up to 20 characters) and an https buttonUrl to send a link button; the message is then limited to 640 characters.
Private reply to comment — send one DM to a commenter. commentId is blank by default after Instagram comment received. Same body and button rules as above. Instagram allows one private reply per comment, usually within 7 days.
Reply to Instagram comment — post a public reply under a comment (up to 2,200 characters). commentId defaults to the triggering comment.
All three take an optional instagramConnectionId. Leave it blank to use the account from the trigger, or the workspace's only connected account. A workspace with several accounts and a non-Instagram trigger must pick one. Output is instagramConnectionId plus messageId (DMs) or id (public replies). A dry run validates the fields without sending.
Send email / Send email batch — a suppressed address, an opted-out contact, or marketing mail with no matching contact is skipped, not failed. Send email returns skipped: true with a reason. Batch keeps going and returns a skipped count plus skippedReasons.
Record bounce or complaint — feed your email provider's bounce and spam-report webhooks back in. Use an inbound webhook trigger, then set feedbackType (bounce or complaint) and the recipient email or the send's messageId. The latest matching send is marked bounced or complained, and that address is skipped from then on. A complaint also opts the contact out of marketing. Output: matched, deliveryId, alreadyRecorded, optedOutContactIds.
Log — writes a line to the run log. Useful while testing. Do not log secret().
HTTP Request — GET, POST, PUT, PATCH, or DELETE. URL and body are templates. Default fail on error fails the node on a non-2xx response. Attach a bearer / API-key / header secret on the node, or interpolate {{ secret("NAME") }}.
HTTPS in production. Private, loopback, and link-local hosts are blocked (same SSRF rules as outbound webhooks).
AI Agent
Drag AI Agent onto the canvas. Connect an OpenAI or Anthropic node to the bottom model handle. The provider node needs a matching secret (Workflows → Secrets) and a selected model. Optional tool nodes hang off the bottom tool handle.
The agent prompts are templates. Default user prompt is {{ input }}.
Plain text output:
outputs["AI Agent"].text;
outputs["AI Agent"].usage.totalTokens;Structured — set Reply as to Structured, then add fields (name + type). Switch the editor to JSON to see or paste the same map (or a JSON Schema object with properties). Switching back to Fields converts it. A choice field is a comma-separated list of allowed values. Then read outputs["AI Agent"].object.sentiment. With no fields, the reply is stored as object.result.
Tools
- HTTP request — the agent can call a URL (same SSRF rules as HTTP Request)
- Set variable — writes
vars.<name>for nodes after the agent - Run workflow — starts another workflow (nested automations stop at depth 2)
- Planeward — one tool per action (create/update/get task, customer, contact, form, pipeline, opportunity, invoice, payment; get submission; finalize invoice; channel message). Inspector values are defaults; the model can override them. These call the same workspace services as the action nodes — they do not go through Planeward’s inbound
/api/mcp/{workspaceId}. - MCP — call tools on a remote Streamable HTTP server. Pick a saved connection from Developer → MCP (header, bearer, API key, or OAuth) or paste an inline URL plus a header/bearer/API key secret. OAuth is only on saved connections. An empty allowlist exposes every remote tool.
Tools still run when Reply as is Structured. The provider and tool nodes do not run as their own steps. The agent calls the model and tools during that one node.
Do not point an MCP tool at this workspace’s /api/mcp endpoint. Use the Planeward tools for workspace actions.
Inbound webhook → classify → create a task. Trigger: inbound webhook. AI Agent (Structured) user prompt:
Classify this request as urgent or normal: {{ input.body.message }}Add fields: priority (choice: urgent, normal) and title (text).
If: outputs["AI Agent"].object.priority === "urgent". True path: Create task title {{ outputs["AI Agent"].object.title }}.
HTTP output:
outputs["HTTP Request"].ok;
outputs["HTTP Request"].status;
outputs["HTTP Request"].body;Secrets
Workflows → Secrets stores values for HTTP, webhook, and AI provider nodes. You see the value once. Types include secret, header, bearer token, API key, OpenAI, and Anthropic. Attach OpenAI / Anthropic secrets on the model node under an AI Agent (they still work as HTTP credentials too).
Prefer the credential picker on HTTP Request, inbound-webhook header auth, or the AI provider node. secret("NAME") in a template is for when you must interpolate. Runs redact known secret strings in stored logs. Still do not put keys in Log nodes. Details: Expressions.
Runs
Executions lists each run. Open one to see the graph as it was, node results, logs, and Retry if it failed.
| You’ll see | Meaning |
|---|---|
ExpressionError | Bad path, unknown identifier, blocked key, empty expression |
HTTP 4xx/5xx | Request failed (failOnError is on) |
| Permission error | The workspace cannot create or view that record |
| Cancelled / timed out | Hit the run timeout |
A failed node stops only the nodes after it. Other branches from the same node keep running, and the run is marked Failed. A node that two branches lead to still runs if one of them reaches it.
Retry replays the same trigger payload. Retry from failed node keeps the results of the nodes that succeeded and runs every failed node again, then the nodes after it.
Settings
Open Settings on a workflow to rename it or edit the description. The danger zone can Duplicate (new draft), Activate / Pause, or delete.
Limits
- Run timeout is about 30 seconds of work, including short Wait sleeps. A durable Wait starts a fresh 30 seconds for the nodes after it. Settings does not expose a timeout control today.
- A run stops after about 1,000 node executions (Foreach counts).
- An expression is at most 2 KB and 256 AST nodes.
- Nested automations stop at depth 2. A workflow that creates a task will not keep chaining
task.createdforever. - Inbound webhook path is a UUID. JSON body only.
Ideas
- Invoice overdue → get invoice → if overdue → post in a channel
- Inbound webhook → create customer → create invoice
- Form submitted → get submission → if a contact was created → channel message
- Task created → if title includes
urgent→ channel message - Webhook
input.body.items→ Foreach → create one task per item - Schedule → HTTP request → log the response
- Inbound webhook → AI Agent (Structured) → if urgent → create a task
Planeward Docs