> ## Documentation Index > Fetch the complete documentation index at: https://hellotars.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Create a voice agent > Create a voice agent, write its prompt and opening line, and set its voice, language, model and call settings. > [!NOTE] > > This capability is in **beta**. Behavior can change before general availability. At the end of this page, you have a voice agent with a prompt, an opening line, a voice and the settings each call uses. To build the call as flow nodes instead of one prompt, continue with [Build a voice flow](https://hellotars.com/docs/guides/voice/build/build-a-voice-flow.md). ## Before you start Before you start, sign in to [the dashboard](https://dashboard.hellotars.com) and select the organization you want to work in. * Voice is on for your organization. See [Who can use voice agents](https://hellotars.com/docs/guides/voice/voice-overview.md#who-can-use-voice-agents). * Your role includes **Build chat agents, voice agents, knowledge bases & tools**. ## Create the voice agent **1. Open Voice Agents in the sidebar and select New voice agent** The **New voice agent** dialog opens. **2. Enter a Name** The name shows in the voice agent list and in call history. **3. Choose the Type** **Single prompt** is one prompt for simple, open conversations. **Conversation flow** uses flow nodes and transitions for structured tasks. **4. Choose the Language and the transcription Model** Flux is selected by default when it supports the language. Otherwise, Nova-3 General is selected when it supports the language. **5. Choose a Voice** The voice is what callers hear. You can change it later. **6. Optional: describe the task in What should this agent do?** Tars turns this description into the first draft of the voice agent's instructions. The limit is 500 characters. **7. Select Create** The voice agent opens on its **Agent** page as a draft. If your plan has no room for another voice agent, the dialog says so. Archive a voice agent you no longer use, or upgrade your plan. ### Import a voice agent instead **Import agent…** on the **Voice Agents** page creates a voice agent from a Tars agent JSON export. Choose the **Agent file**, enter a **Name**, then select **Import agent**. Review the imported settings in the voice agent before you publish it. To load an export into an existing voice agent, open it and select **Import JSON…** from the more actions menu. Tools and knowledge bases are matched by name. Anything that does not exist in your organization is flagged so you can fix it after the import. The import either replaces the selected draft or creates a new draft, as you choose. Live calls are untouched until you publish. ## Write the prompt A single prompt voice agent opens on its **Agent prompt**. Write who the voice agent is, what it must do, and what it must never do. Select **Insert variable** or **Insert tool** to place a variable or a tool name in the prompt. [https://dashboard.hellotars.com](https://dashboard.hellotars.com) ![The Agent page of a single prompt voice agent with the model, voice, language and Handbook toolbar and the prompt editor highlighted, and the settings rail on the right](https://hellotars.com/docs/docs/guides/voice/images/voice-agent-page.png) When the voice agent follows steps, split the prompt into conversation states: * A conversation state is one step of the call, with its own instructions. * The **Starting state** is where a new call enters. * A transition moves the caller to another state when its condition is met. Select **Add transition** on a state, then describe the condition in **When should this happen?**. * **Capture when moving** records values from what the caller already said when the call takes a transition. If a required value is missing or not valid, the call stays in the current state. ## Set the opening line The **Welcome message** decides how the call starts. | Setting | What it does | | - | - | | **Who speaks first** | **Agent speaks first** or **Caller speaks first**. | | **Opening line** | What the voice agent says first. **Fixed message** says your exact text. | | **Pause before speaking** | The silence after the call connects, before the opening line. | | **Agent starts speaking after silence** | When the caller speaks first but says nothing, the voice agent opens after this time. | ## Set the voice, language and model The toolbar above the prompt holds these controls: | Control | What it does | | - | - | | **Model** | The language model. **Model settings** holds **Temperature** (lower is more predictable, 0.3 is recommended for voice), **Reasoning effort** for models that reason, and **Structured tool calls**. | | **Voice** | The voice the caller hears. | | **Language** | Up to 10 speech languages. The first is the primary language. The voice agent answers in whichever language the caller uses. | | **Handbook** | The **Agent Handbook**: built-in guidelines for **Personality & Tone**, **Accuracy & Format** and **Trust & Safety**, such as **Echo verification** and **Scope boundaries**. | | **Current time awareness** | The voice agent's timezone, so it reads "today", "tomorrow" and business hours correctly. | With a speech-to-speech model, **Fall back to a standard model if speech-to-speech fails** keeps the call going on your saved standard model and a compatible voice. ### Update the Handbook preset text Each voice agent keeps the version of the Handbook text it started with. A new voice agent starts on the latest version. An existing voice agent stays on its version until you update it and publish. **1. Select Handbook in the toolbar** The footer shows **Preset text version** and the version number. **2. Select Update preset text** The button shows only when a newer version exists. The update changes your draft, not the live version. **3. Publish the voice agent** New calls use the new text. See [Publish a version](https://hellotars.com/docs/guides/voice/launch/test-and-publish-a-voice-agent.md#publish-a-version). ## Configure the settings The settings rail on the right of the **Agent** page groups every other setting. | Section | What it holds | | - | - | | **Speech settings** | The **Voice** and its **Pronunciation** list, a **Background** sound, **Turn-taking** (**Wait before replying**, **Interruption sensitivity**, **Backchannel**), and **While the caller waits** (filler phrases, reminders for a silent caller, and holding lines). | | **Transcription** | The transcription **Model**, **Noise removal**, **Transcription mode**, **Boosted keywords**, and end-of-turn timing. | | **Call settings** | **Web calls**, **Outbound calls** with voicemail and phone menu detection, **Keypad** input, and **Limits** such as **End the call after silence of** and **Maximum call length**. **Access** sets the members who can handle or listen to calls. | | **Post-call extraction** | The fields the analysis model extracts from each call, and whether to **Save caller details** to the end user's profile. | | **Security & storage** | **What to keep from each call**, **Personal data to mask**, **Guardrails**, recording, expiring links for recordings, the **Fallback voice if the main voice fails**, and **Call variables** defaults. | | **Webhooks** | The webhook destinations that receive this voice agent's events. Destinations live in **Settings** → **Integrations** → **Webhooks**. | **Interruption sensitivity** at 0 means the caller never interrupts by speaking. Keypad presses can still interrupt. ### Fallback voice In **Security & storage** → **Defaults**, **Fallback voice if the main voice fails** sets the voice a call uses when the main voice is not available. Choose **Pick automatically** or **Choose myself**. With **Fallback from another provider**, the fallback voice comes from a different voice provider. ### Members who handle calls Under **Call settings** → **Access**, choose the members who can handle or listen to this voice agent's calls. Only members with live chat access can be chosen. An empty selection gives nobody access. Saving the list applies at once and does not publish a version. ### What each call keeps **What to keep from each call** has three options: | Option | What Tars keeps | | - | - | | **Everything** | Recording, transcript, variables, function data. | | **Everything except personal data** | Personal details are masked in text. Original recordings are kept and are not scrubbed. | | **Basic details only** | Duration and outcome. No transcript or recording. | Your organization's **Voice calls** policy sets the most any voice agent may keep. A voice agent can keep less. See [Set the voice call policy](https://hellotars.com/docs/guides/security/voice-privacy.md). When **Record calls** is on, **Tell phone callers** sets how phone callers hear about it. The options are **Nothing**, **Announce once**, **Ask to say yes** and **Ask to press 1**. With **End phone calls if the caller refuses**, the call ends when the caller does not agree. The notice and consent for web callers are set in **Design** → **Call screen**. ## Verify The voice agent shows in the **Voice Agents** list with the status **Draft only**. Open it and select **Test** to talk to it. See [Test and publish a voice agent](https://hellotars.com/docs/guides/voice/launch/test-and-publish-a-voice-agent.md). ## Related pages * [Voice agents](https://hellotars.com/docs/guides/voice/voice-overview.md) * [Build a voice flow](https://hellotars.com/docs/guides/voice/build/build-a-voice-flow.md) * [Give a voice agent tools](https://hellotars.com/docs/guides/voice/build/voice-agent-tools.md) * [Transfer calls](https://hellotars.com/docs/guides/voice/build/transfer-calls.md) * [Set the voice call policy](https://hellotars.com/docs/guides/security/voice-privacy.md) This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.