Operations agent: designing and testing

The Workflow page is the designer: the palette of triggers and steps on the left, the board in the middle, the inspector on the right. Open in a new window gives the board the whole screen.

Start from a template#

Templates opens 18 ready boards, grouped by Yeastar PBX, HubSpot and general use. Every one is tested end to end. Some examples:

• Email the transcript, or a summary, of every call
• Answer your questions about every call in an Excel report
• Post inbound-call transcripts to your system
• Alert the team about missed calls
• Hand every call to several agents at once
• Log every PBX call in HubSpot with a summary
• Every weekday morning, call your endpoint and report
• Archive every recording to your own S3 storage

A template lists the values you must fill in (recipients, addresses). Use this template replaces the board; Use and try it also opens a test run. The board you had is kept in the history, so nothing is lost.

Design with AI#

Design with AI takes what should happen, in your own words, for example "when a call ends, transcribe it, and if the customer asked for a callback email sales". The designer draws it from the steps this platform has, checks that every connection carries the right kind of data, and puts it on the board unsaved for you to review. Tick Change the current board to modify the board you have instead of starting a new one.

Build and edit by hand#

• Connect two steps by dragging from the right handle of one to the left handle of the next. Click an arrow to select it; Delete connection removes just that arrow.
• Undo and Redo (Ctrl+Z, Ctrl+Y), and Delete or Backspace removes the selected step or connection.
• Arrange lines the boxes up horizontally or vertically; Fit to view brings the whole board into view.
• Saving checks the board: a step nothing reaches, a value taken from a step that does not run before it, the wrong kind of data (a web link where a file is expected), or a condition whose branch goes nowhere is refused with a message saying which step to fix.

Test the whole flow#

Test run starts a run with trigger data you type (JSON). With several triggers on the board, choose which one it enters through, or use Run through this trigger in a trigger's inspector. The quickest way to get real data is Listen (see Triggers): capture a real event, then Use as test payload. After a run, click any step to see, in full, the data it received and produced.

Test one step#

The Send email, Webhook, S3 and AWS S3 steps have a panel to preview and test the step alone, without running the rest of the board. The values are filled from the last run, or from the sample data of the Test run dialog.

• Preview shows exactly what would go out: the email as the recipient will see it, or the request with its URL, headers and body. Credentials are hidden.
• Send a test (or Run this step now) does it for real and shows what the other side answered.

Use it to get a webhook body or an email right before you enable the agent.

History, restore, export and import#

Every save is kept in the History, with where it came from (created, saved, restored, imported, AI, template). Compare any entry with the board as it is now, or two entries with each other, and Restore one: restoring never deletes anything, it adds a new entry.

Export / Import saves the board as a JSON file and loads one back, for example to copy a flow to another project. Import checks the file first and shows the current board beside the one from the file before you replace it.

Before you enable#

1. Save, and fix anything the check reports.
2. Run a test with real data from Listen and open the run.
3. Preview and test the deliver steps.
4. Enable the agent under Settings or on the board.