Procedure Editor

The editor

The editor is where you build your procedure as a visual graph. Three panels:

Panel	Where	What it's for
Node list	Left sidebar	Browse and add node types
Canvas	Center	Your procedure graph — nodes and connections
Inspector	Right sidebar	Configure the selected node

Open any procedure from the dashboard, or create one with New Procedure.

---

The canvas

The canvas is your graph. Nodes are boxes; edges (arrows) connect them and define execution order.

• Pan — drag the canvas background
• Zoom — scroll wheel or pinch
• Select a node — click it to open its config in the inspector
• Move a node — drag it
• Delete a node — select it and press Backspace or Delete

Every procedure has a Start node (the entry point) and one or more End nodes. The AI agent begins at Start and follows edges until it reaches an End.

---

Node types

Tool Call

Tool Call nodes are where your procedure talks to the outside world — Stripe, Plaid, Slack, Onfido, your own internal APIs.

Configure in the inspector:

Field	What to set
Connection	Which authenticated account to use (e.g. "Stripe Sandbox")
Action	The specific operation (e.g. stripe.createPaymentIntent)
Inputs	Where each parameter comes from

The result of the call is available to later nodes as {{steps.nodeName.result}}.

Example — charging a customer:

Parameter	Value
amount	{{inputs.amount}}
currency	{{inputs.currency}}
customer	{{inputs.stripeCustomerId}}

You can also mark a Tool Call with Require approval before calling — this pauses the run before the API call happens and shows the reviewer the exact parameters. Useful for wire transfers or other high-impact operations.

---

Human Review

Human Review nodes pause execution and wait for a team member to approve or reject.

Configure in the inspector:

Field	What to set
Instructions	Plain text for the reviewer — what to check, what criteria matter

When the run reaches this node:

1. Status changes to awaiting_review
2. Run appears in the Pending Reviews queue
3. Execution pauses until a decision is made
4. Approved → execution continues; Rejected → run fails with the reason logged

→ See Human Review for the full guide.

---

Condition

Condition nodes branch the graph based on a boolean expression.

Configure in the inspector:

Field	What to set
Expression	A boolean expression referencing inputs or step results
True branch	The edge to follow when the expression is true
False branch	The edge to follow when the expression is false

Example expressions:

{{inputs.amount}} > 10000

{{steps.kycCheck.result.status}} === "approved"

{{steps.balanceCheck.result.available}} >= {{inputs.transferAmount}}

Expressions are evaluated as deterministic code — not by the AI. The result is always the same for the same inputs. This matters for financial thresholds and credit decisions.

---

Wait

Wait nodes pause execution for a set duration.

Field	What to set
Duration	How long to wait — e.g. 30m, 24h, 7d

Useful for multi-step workflows that span hours or days: wait 24 hours after sending an onboarding email, then check whether the customer completed verification.

---

Connecting nodes

To connect two nodes:

1. Hover over the source node until its output port appears (a small circle on the right or bottom)
2. Click and drag from the port
3. Drop onto the input port of the destination node

Delete an edge by clicking it and pressing Backspace.

Condition nodes have two output ports — true and false — each must connect to a separate downstream node.

---

Input mappings

Each node's parameters can reference data from earlier in the procedure:

Reference	Where the data comes from
{{inputs.fieldName}}	The procedure's input schema
{{steps.nodeName.result}}	The full result of a prior step
{{steps.nodeName.result.fieldName}}	A specific field from a prior step

The inspector provides autocomplete for available references. For example, after a Stripe customer lookup, you can type {{steps. and see steps.lookupCustomer.result.email suggested.

---

Saving and publishing

Action	What it does
Save	Saves the current state as a draft. Doesn't affect live runs or triggers.
Publish	Makes this the default version. All unpinned triggers and API callers use it from now on.

Save frequently. Publish when you've tested and are ready to ship. Published versions are immutable.

→ See Versions for the full versioning guide.

---

Simulation mode

Before using real integrations, click Simulate in the toolbar to do a dry-run:

• The graph executes using mock responses you configure per node
• Conditions are evaluated with your simulated data
• The step log is shown in the editor
• No real API calls are made

Simulation is the fastest way to check that input mappings resolve correctly and conditions route as expected — before connecting anything real.