# Voice prompt

> Ask your AI assistant to edit your AI agent's call script, voice, language model and conversation timing, and to manage your saved prompts.

- Section: Using the MCP (AI assistant) › AI agent setup
- Web page: https://docs.biglysales.com/mcp/voice-prompt
- Manual (in-app) equivalent: [Voice prompt](https://docs.biglysales.com/manual/voice-prompt.md)
- Video: placeholder — "Editing the voice prompt and voice from your AI assistant" (not yet recorded)

The **voice prompt** is the script and set of rules your AI agent follows on every call. It also holds the agent's voice, the language model that does the thinking, the speech recognition model, and timing options such as how long a call may last and how long to wait in silence.

It's the same setup you see on **AI › Teach Your Bot › Voice Prompt** in the app. Your assistant can read it, change any part of it, and keep named copies of your prompts.

> **Example request to your AI assistant:** "Show me our outbound prompt and the voice we use."
> "Switch to a calm female English voice."
> "Rewrite the objection-handling section of our outbound prompt to be shorter."
> "Save the current prompts as Spring campaign."

## Before you start

- [Connect your AI assistant](https://docs.biglysales.com/mcp/connect.md) to Bigly.
- You need permission to **read** the voice prompt to see it, and permission to **update** it to make changes. Account owners and admins have both by default.
- Your prompt must contain `{intro}`, so your [intros](https://docs.biglysales.com/mcp/intros.md) are folded in. The assistant can check this for you.

> **Let the assistant follow the writing guide:** Bigly publishes a prompt-writing guide that your assistant can load. Ask it to *follow the Bigly prompting guide* when it writes or rewrites a prompt. See [Reference resources](https://docs.biglysales.com/mcp/reference-resources.md).

## What the assistant reads

- Every setting below, with its current value.
- The **voice, language model and speech-to-text model** you've selected, by name.
- The **catalog** of every voice and model you can choose from, with each voice's language and gender. Describe what you want ("a warm male Spanish voice") and the assistant picks a match from the catalog.
- Your **saved prompts**, newest first.

## Change the prompt or settings

1. Describe the change.
2. The assistant shows a **preview** listing each field that changes, with its current value and its new value. For long prompts, ask it to show you the full new text.
3. Confirm in the chat. The change applies to the next call.

If you edit a prompt that was loaded from a saved prompt, the saved copy is updated too.

## Settings explained

| Setting | On screen | Accepted values |
|---|---|---|
| Outbound and inbound prompt (`outbound_prompt`, `inbound_prompt`) | **System Prompt** | Text |
| Prompt prefix (`prefix`) | | Text added before the prompt |
| First and second outbound sentence, first inbound sentence (`outbound_first_message`, `outbound_second_message`, `inbound_first_message`) | **First sentence outbound calls**, **Second sentence outbound calls**, **First sentence inbound calls** | Text |
| Goal of the call (`prompt_type`) | **Type** | Transfer call or meeting scheduler |
| Language model (`ai_large_language_model_id`) | **Model** | One from the catalog |
| Voice (`ai_voice_id`) | **Voice** | One from the catalog |
| Speech-to-text model (`ai_speech_to_text_model_id`) | **Speech to Text** | One from the catalog |
| Max call length (`max_duration_of_call`) | **Max Duration Of Call** | 30 to 1,800 seconds |
| Max ringing time (`max_ringing_time`) | **Max Ringing Time** | 5 to 99 seconds |
| Silence timeout (`silence_timeout`) | **Silence Timeout** | 1 to 300 seconds |
| Speak-up delay (`idle_user_speak_duration`) | | 1 to 15 seconds before the agent speaks to a silent caller |
| Creativity (`temperature`) | | 0 to 2 |
| Top P (`top_p`) | | 0 to 1 |
| Max tokens (`max_tokens`) | **Max Tokens** | 1 or more |
| Allow interruptions (`allow_interruption`) | **Interruption** | On or off |
| Interrupt Min Words, Interrupt Speech Duration | **Interruption** | 0 or more |
| Min and Max End Pointing Delay, minimum silence duration | **Voice Activity Detection** | 0 or more |
| Voice language (`voice_language`) | **Language** | en or es |
| Volume (`voice_volume`) | **Volume** | 0.01 to 1 |
| Speed (`voice_speed`) | **Speed** | 0 or more |
| Emotion (`voice_emotion`) | **Emotion** | One of the emotions listed in the app, such as neutral, calm, confident or happy |
| Speech language (`speech_language`) | **Language** under Speech to Text | en-US, es-419 or multi |
| Key Terms (`key_terms`) | **Key Terms** | Comma-separated words the speech recognition should listen for |
| Use Context (`use_context`) | **Use Context** | On or off |
| Turn Detection (`turn_detection_provider`, `turn_detection_language`) | **Turn Detection** | English or multilingual |
| Background Sound (`background_audio_file`, `background_audio_volume`) | **Background Sound** | A file name, and a volume from 0 to 1 |

Keep the voice language and speech language consistent with the voice you choose. A Spanish voice with English speech recognition gives poor results.

## Saved prompts

Saved prompts are named copies of your outbound and inbound prompts.

- **List** them by asking *show our saved prompts*.
- **Save** the current or a new prompt under a name. A name is required.
- **Delete** one by name. The assistant asks you to confirm first.

If the saved prompt you delete is the one your agent is using, its outbound and inbound prompts are emptied too. Your agent then has no script until you set one. Ask the assistant whether a saved prompt is in use before deleting it.

## Limits and rules

- Only the settings in the table can be changed. Anything else is rejected.
- Numbers must be inside the ranges shown.
- Voices and models must be chosen from the catalog. An unknown ID is rejected.
- Placeholders in the prompt, like `{company_name}`, are not checked. Use a [call preview](https://docs.biglysales.com/mcp/call-context-preview.md) to spot ones that don't resolve.

## What you can't change here

- **Livekit Version** and **Sip Provider** are account settings. See [AI settings](https://docs.biglysales.com/mcp/ai-settings.md).
- **Tools** the agent can use, such as transfer and end call, are set in [Tools configuration](https://docs.biglysales.com/mcp/tools-configuration.md).
- The Bigly **Prompt Library** (the curated prompts you pick on the Prompt Library tab) is managed in the app only. See [Prompt selector](https://docs.biglysales.com/manual/prompt-selector.md).
- **Test Call** can't be placed from the MCP. Use a [call preview](https://docs.biglysales.com/mcp/call-context-preview.md) to see the assembled script, or place a [test call in the app](https://docs.biglysales.com/manual/test-calls.md).

## Related

- [Voice prompt in the app](https://docs.biglysales.com/manual/voice-prompt.md)
- [Intros](https://docs.biglysales.com/mcp/intros.md)
- [Prompt variables](https://docs.biglysales.com/mcp/prompt-variables.md)
- [Guided workflows](https://docs.biglysales.com/mcp/guided-workflows.md), including *change voice or model*

---

All documentation pages: https://docs.biglysales.com/llms.txt