Test your agent with real calls, use the Customize Behavior chatbox to tweak instructions, and publish when it's ready! ๐งชโ
The Customize Behavior chatbox lets you change your agent's instructions using plain English โ no need to edit the prompt by hand!
After each change, the chatbox might suggest more tweaks. Click a suggestion to apply it right away!
You can upload documents to help the chatbox write better instructions:
| Supported Formats | Max Size |
|---|---|
| PDF, CSV, XLS, XLSX, DOC, DOCX, JPG, JPEG, PNG | 10 MB per file |
Upload up to 20 documents. The AI reads the text from them and uses it to make smarter prompt changes. ๐
When you're building a new agent (before saving it), the chatbox works in draft mode. Your instructions and chat are saved to a temporary session. Once you save the agent, everything moves over to the saved version.
Try out your agent with real phone calls before going live!
The panel shows what's happening in real time:
| Status | Description |
|---|---|
| Registered | The call is being set up โณ |
| Ongoing | The call is happening right now ๐ด |
| Ended | The call is over โ |
When the call ends, you'll see:
| Data | Description |
|---|---|
| Call Summary | A short recap of the conversation written by AI ๐ |
| Sentiment | How the caller was feeling (positive, neutral, negative) ๐๐๐ |
| Success | Did the call go well? |
| Voicemail Detection | Was a voicemail found? |
| Saved Fields | Values extracted by post-call analysis (e.g. intake answers, custom fields) |
| Transcript | The full word-for-word conversation |
| Disconnection Reason | Why the call ended |
If your prompt uses placeholders, open the Variables tab before testing to pre-fill values. This shows how the agent behaves with real client data.
Variables are displayed with human-readable labels derived from your form's task titles and field definitions. For example, a variable like t1_value shows up as the task's actual question name rather than a raw key.
Auto-managed variables
Some variables โ such as task statuses (t1_case_task_status) โ are pre-filled with a default value (e.g. todo) and shown as a dropdown so you can test different status paths without editing the prompt manually.
Variable values are saved per agent in your browser, so they persist between sessions and reopen automatically the next time you run a test call.
Every time you publish, a full version snapshot is saved. This captures all your agent's settings โ form, phone number, transfer, email, SMS, welcome messages โ so you can restore any past version completely.
| Field | Description |
|---|---|
| Version Number | A number that goes up with each publish |
| Version Title | An optional short label you set when publishing |
| Version Description | An optional note about what changed in this version |
| Agent Name | The agent's name at that time |
| Last Modified | When this version was last changed ๐ |
| Status | Published or Draft |
When you load a past version, the editor header shows a grey Viewing V{n} badge with the version title. This is a read-only preview โ your live agent is unchanged until you explicitly publish again.
Publishing makes your agent's current settings go live. All calls handled by this agent will use the published version.
When you click Save & Publish, a dialog appears before the agent goes live:
| Field | Description |
|---|---|
| Version Title | A short label for this version (e.g. "Added Spanish support") โ shown in version history |
| Version Description | A longer note about what changed โ shown in the editor header and history sidebar |
Both fields are optional. Skip them and click Publish to publish without a label.
WARNING
Publishing happens right away! Once you publish, all active calls will use the new settings. Always test well before publishing. โ ๏ธ