# Voice Autopilot connection setup

Voice Autopilot connects incoming telephone calls to a realtime AI voice agent with configured instructions, tools and human hand-off.

*Release 3.3 · admin, operator, visitor*

**Settings › AI › Features** · **Settings › Voice › Voice Numbers** · **Settings › AI › Accounts**

## What it is

**Voice Autopilot** lets an AI agent listen and reply during an incoming telephone call. It uses the selected voice, speech model and instructions. The agent can use enabled actions and hand the call to a human when needed.

## Who uses it and why

Administrators configure the voice feature and connect it to a telephone number. Operators receive calls handed to the human queue and can use available transcript text for context. Callers speak to the agent through their telephone.

## Where to find it

Create or edit **Voice Autopilot** in **Settings › AI › Features**, using the **Settings** role with **AI** access or the **Developer** role.

With the **Settings** role, open **Settings › Voice › Voice Numbers**, edit a number and use its **AI** tab to select **Voice AI Feature**. Powerfront manages the underlying account in **Settings › AI › Accounts**.

## How it works

1. A call reaches a number configured for Voice Autopilot.
2. INSIDE connects telephone audio to the configured realtime speech model and waits for confirmation of the session settings.
3. The agent follows its **Rules**, listens to the caller and replies using **Voice**.
4. Enabled actions can check availability, transfer to an operator, request a callback, prepare voicemail or end the call. Configured custom functions can provide additional actions.
5. On human hand-off, INSIDE moves the voice conversation to the queue. It prevents competing terminal actions from running at the same time.

The service records configured and applied speech settings for diagnosis, including an internal conversation comment at the end of the call. Caller transcription depends on **Transcription Model** being set.

## Configure it

Ask Powerfront to prepare a suitable AI account and OpenAI credential and confirm that telephone service is enabled. Create an enabled **Voice Autopilot** feature with suitable **Rules** and actions, then select it under **Voice AI Feature** on the number.

Choose **Model**, **Voice** and **Max Duration (secs)**. The duration defaults to 600 seconds, with an editor range of 60 to 3,600 seconds. Set transcription options for the languages and terminology your callers use.

Configure greeting and hand-off wording and **VAD Mode** as described in the voice turn-taking article. Verify a test call, human hand-off and any callback or voicemail actions you enable.

### Settings

| Setting | Where | Applies to | Default | Who can change it | What it does |
|---|---|---|---|---|---|
| Voice AI Feature | Settings › Voice › Voice Numbers | channel | None (Disable Voice AI) | Settings role | Assigns an enabled Voice Autopilot feature to the telephone number. |
| Model | Settings › AI › Features | website | gpt-realtime | Settings role, Developer role | Selects the realtime speech model used for the voice session. |
| Voice | Settings › AI › Features | website | alloy | Settings role, Developer role | Selects the speaking voice callers hear. |
| Max Duration (secs) | Settings › AI › Features | website | 600 seconds | Settings role, Developer role | Limits AI call handling before transfer to the human queue. |
| Transcription Model | Settings › AI › Features | website | gpt-4o-transcribe | Settings role, Developer role | Selects caller speech transcription; an empty value disables transcription. |
| Transcription Language | Settings › AI › Features | website | Auto Detect | Settings role, Developer role | Uses automatic language detection when unset, or restricts transcription to the selected language. |
| Transcription Prompt | Settings › AI › Features | website | Not set: no additional transcription guidance | Settings role, Developer role | Supplies terminology or context to help transcribe caller speech. |
| Rules | Settings › AI › Features | website | Not set: missing instructions cause hand-off to the queue | Settings role, Developer role | Supplies instructions for how the voice agent should handle calls; missing instructions cause hand-off to the queue. |

### Permissions

| Permission | Lets you |
|---|---|
| AI | With the Settings role, allows Voice Autopilot feature configuration; the Developer role can also access Features. |

## Limits and good to know

- This release's realtime voice integration uses OpenAI Realtime and Twilio telephone media.
- Missing AI instructions cause hand-off to the queue. Provider errors and maximum duration can also trigger hand-off; queue availability still follows your voice setup.
- Session readiness has a 10-second deadline. The service also detects a response stalled without incoming audio or events for 10 seconds while awaiting output.
- An empty **Transcription Model** disables caller transcription. Audio conversation does not by itself establish a complete text transcript.
- Changing the text-model provider on an account does not establish support for another realtime voice provider.
- Recording and retention follow the separate telephone settings.

## Depends on and works with

### This feature uses

- **Requires** [AI accounts and shared defaults](AIP-001.md): Powerfront prepares the AI account and OpenAI credential used by the realtime connection.
- **Requires** [Voice provider accounts and telephone numbers](RTC-001.md): An enabled voice account and telephone number with Voice AI Feature selected are needed to receive calls.
- **Configured in** [AI feature settings and change history](AI-001.md): Feature availability and saved configuration determine the voice agent used by a number.
- **Configured in** [Voice Autopilot call handling](AI-013.md): Voice Autopilot instructions and enabled actions govern the conversation.
- **Configured in** [AI functions and required inputs](AI-016.md): The selected AI functions determine which call actions and custom tools the voice agent may use.
- **Configured in** [Voice Autopilot turn detection and interruptions](AIP-026.md): Turn detection, greetings and playback handling govern spoken interaction.
- **Works better with** [Voice queue and call offers](RTC-003.md): Human hand-off places the conversation in the voice queue.
- **Works better with** [Telephone callback requests](RTC-009.md): An enabled callback action can register a telephone callback.
- **Works better with** [Voicemail messages and follow-up](RTC-010.md): An enabled voicemail action can redirect the caller to recording.
- **Related** [Telephone call recovery and wrap-up](RTC-015.md): Use telephone call diagnostics to investigate setup or human hand-off failures.

### Used by

- **Required by** [Voice Autopilot call handling](AI-013.md): Realtime voice AI provides speech processing for the call.
- **Required by** [Voice Autopilot turn detection and interruptions](AIP-026.md): Turn detection and playback synchronisation operate within a realtime Voice Autopilot session.

## FAQ and troubleshooting

### Why can I not select Voice Autopilot as a feature type?

The website needs telephone service enabled. Ask Powerfront to confirm your voice setup.

### Why did the call go to a human immediately?

Ask Powerfront to check the selected feature, its **Rules**, credentials and realtime connection. Missing instructions and connection failures can trigger queue hand-off.

### Why are caller words missing from the transcript?

Check **Transcription Model** and **Transcription Language**. An empty model disables transcription; incorrect language or audio quality can also affect results.

### Why was the caller transferred after ten minutes?

The default **Max Duration (secs)** is 600 seconds. Review that setting and **Max Duration Transfer Message** for the feature.