Instructions

Define one clear Conversation objective, then write the instructions the agent needs to achieve it. The prompt supports Markdown and up to 28,000 characters. Generate with AI drafts instructions from the objective and agent configuration; review the result before publishing. Copy prompt copies the full text.

Choose how calls begin

Mode	Behavior
Wait for the caller (`user_initiates`)	Waits for the other person to speak.
Agent greets in its own words (`ai_dynamic`)	Speaks first using the prompt.
Agent reads a fixed greeting (`ai_defined`)	Reads the configured opening phrase.

The opening delay is in Advanced → Conversation, from 0 to 5 seconds. Switching modes preserves the fixed phrase for later use.

Work in the editor

The toolbar supports headings, lists, bold, italic, inline code, search, and a table of contents. Search uses Ctrl+F; Enter and Shift+Enter move between matches. Select text for the floating formatting toolbar. Type @ at the start of a word to select a tool already attached to the agent. The editor inserts its exact name in backticks. A mention does not create or connect a tool. Use paired markers on separate lines for frequently maintained sections:

[[Business hours]]
Monday through Friday, nine in the morning to six in the evening.
[[/Business hours]]

The Editable sections panel lets your team update these blocks. Titles cannot contain brackets or newlines or start with /. Sections cannot nest. Cards and the prompt edit the same text; this is not a separate shared library.

Use variables

Enable the relevant blocks in Advanced → Inbound context. Their content is already included in the final prompt; insert a variable explicitly when you need it in a particular place.

Variable	Content
`{{contact_info}}`, `{{contact_name}}`	Contact information or name.
`{{contact_properties}}`	Selected custom contact properties.
`{{property_name}}`	An individual enabled property; empty when absent.
`{{call_history}}`	Configured recent-call summaries.
`{{last_transcript}}`	Most recent available transcript, truncated to the configured limit.
Inbound webhook keys	Context returned by your endpoint; overlapping keys take precedence.

Test variables can also be provided in the test panel. Always explain what the agent should do when a value is missing. Double braces substitute values; they do not provide a conditional template language. Prompt changes autosave to a draft and take effect after publishing. Write in Spanish or English as appropriate, specify spoken languages in General, and verify the result with representative calls.

Frequently asked questions

Why does the agent still follow old instructions?

Check for save errors and publish the draft before testing.

How do I improve spoken numbers?

Add concrete examples for phone numbers, prices, dates, and abbreviations, then test them aloud. Speech normalization is also available in Advanced.

Should documents go in the prompt?

Keep essential behavior and short reference facts here. Put long documents in a connected knowledge base and explain when to search it.

Why does @ show no tools?

Add tools in the Tools tab first.

Get started

Voice agents

Text agents

Messages

Calls

Contacts

Contact center

Lead routing

Reports

Organization

Billing

Integrations

MCP

Choose how calls begin

Work in the editor

Use variables

Frequently asked questions

See also

​Choose how calls begin

​Work in the editor

​Use variables

​Frequently asked questions

​See also

Choose how calls begin

Work in the editor

Use variables

Frequently asked questions

See also