Skip to main content

The editor: canvas, palette, Begin

Module goal: move around the editor with confidence and understand what each button in the header does.
Agent editor: the left panel is hidden, the canvas in the centre, the node inspector on the right

/editor/‹agent› — three zones: the panel on the left (Nodes · Agent variables · Settings · Issues), the canvas in the centre, the inspector of the selected node on the right. The ◧ and ◨ buttons in the header hide the side panels when you need the whole canvas.

The editor header, in order

The header is a single row:
  • Dialogue / Pipeline — the switch in the centre: the conversation map or the functions before and after the call.
  • Search icon (Find in document) — search across the whole agent document. See below.
  • More (⋯) — Share, Import, Export; import and export exchange the agent as JSON.
  • Save — shown only while there are unsaved edits; it writes a draft. The draft does not affect live calls. Once everything is written, Saved · date takes its place.
  • Publish vN — immediately fixes the draft as a new version; calls run on that version. A version note, the version history and discarding edits are in the menu of the arrow next to it. In detail — Publishing and versions.
  • Test — chat and voice call on the draft or on the published version.
  • Copilot — edits to the flow in words, always through a proposal that you accept.
Issues is not a header button but a tab of the left panel: graph validation; errors block publishing, warnings do not. Organize is among the canvas control buttons, next to zoom and the lock; after it a notification Layout organized appears with an Undo link. The lock locks the canvas: nodes don’t move, edges can’t be drawn, Delete removes nothing. A node is added by clicking its card in the palette (the Nodes tab; the Find a node type field filters the types), an edge by dragging from a node’s bottom port to another node. Clicking a node opens the inspector on the right: everything inside the node is configured there.

Search in the document

In a large agent, the node you need is faster to find by search than by eye. Where to start. The search icon in the editor header, tooltip Find in document; in the component editor the same button is in its header. The Cmd/Ctrl + F shortcut does the same. A floating panel opens above the canvas — it is a panel, not a button: there is no point looking for a button on the canvas itself.
The agent document search panel: results for the query “administrator”, grouped by node

The search panel: the query “administrator”, counter 1/9, results grouped by node — “No slots”, “Asks for a human”, “Transfer to an administrator” and others. The match is underlined both in the list and on the node card on the canvas.

The browser’s search does not work for this: it only sees the short previews on the cards and cannot scroll the canvas to the match. The results are split into three categories, with a tab for each under the query field:
  • Nodes — node names, prompts, transition conditions, authored node fields, then the pipeline functions.
  • Values — the text you typed in the agent’s settings: the global prompt, post-analytics fields (name, extraction instruction, choices — also when analytics is off, marked “off”) and agent variables (name and default value). In the component editor — the component prompt.
  • Settings — the setting itself, found by its label on screen: rows and switches of the Settings tab, sections of the Quality & integrations and Additional settings windows, the answering machine detection and call screening dialogs. Type “answering machine” and the panel shows where that setting lives.
With no tab selected, the panel shows everything. A selected tab keeps only its results, and the counter and stepping run within it. A tab with no matches is inactive. Dictionaries, output variables and internal identifiers are deliberately left out of the index — a JSON fragment in the results is unreadable and may reveal a secret. The unit of navigation is an occurrence, not a card: if one node has three matches, you will step through all three. Enter and Shift+Enter step forward and back, and while the panel is open so do Cmd/Ctrl + G and Shift + Cmd/Ctrl + G, from anywhere. Crossing between dialogue nodes and pipeline functions switches the view by itself. The ceiling is 500 occurrences; if there are more, the panel says it is showing the first ones. A step to a setting or a variable opens the place where it is edited: the left panel tab, or a window such as Agent variables or Quality & integrations. The window opens next to the search panel, so you can keep stepping; Esc closes the panel first. On a narrow screen the panel shrinks to a strip with the query field above the window. The match is highlighted in the field itself, where the text is edited, and the current occurrence scrolls into view.
The shortcut yields to input fields. If the cursor is in a prompt, in any text field or in the code editor, Cmd/Ctrl + F goes to the field — you get the browser’s ordinary search, and the panel does not open. This is intentional: an author editing a long prompt expects to search that prompt. Once the panel is open, the shortcut works from anywhere and puts the cursor back into the query field.Hence the most common complaint, “there is no search”: Cmd/Ctrl + F was pressed without taking focus off a field. Click an empty spot on the canvas — or just press the search icon in the header.
About autosave and “conflict”The draft is saved automatically, with a short delay after the last edit. If you keep typing while it saves, the canvas does not roll back — the server response applies only service data, and your unsaved changes stay in place.The message save conflict: the flow was changed in parallel — reopen the flow means exactly what it says: someone is editing the same agent in another window or as another user. It is worth acting on, not closing out of habit.
Begin node inspector: who speaks first

Begin. The main setting of this node is who speaks first. For outbound dialling — the agent (otherwise both will stay silent). For an inbound line it is usually the agent too: it greets the caller.

Begin node inspector: who speaks first
If the client should speak first but silence on the line is not acceptable, turn on Start if the caller is silent in the User speaks first card and set the seconds next to it: after that much silence the agent speaks on its own. In detail — Begin node.
Advice on the first line. The agent speaks its first line from the node the Begin edge leads to. For an outbound call it is worth naming yourself, the company and the purpose of the call right away — in the client’s language. For example: “Hello, this is Aliya from the … clinic, I’m calling to remind you about your appointment”, or for the US — “Hi, this is Alex from …, I’m calling about your appointment”. This is a recommendation for the node’s text: the platform does not add such a line by itself.
PitfallsBegin cannot be deleted and has exactly one outgoing edge — it sets the entry point into the dialog. If after edits the conversation starts “in the wrong place”, first check where the edge from Begin leads: it is shown in the Begin inspector as First node, and you can change it there.