Skip to main content
Creating an assistant takes about two minutes. You pick a template that matches your use case, give the assistant a name, then fine-tune the model and transcriptor settings for the best call quality. This page covers the full creation flow and both configuration tabs in detail.

​
Create an assistant

1

Open the Assistants page

In the left sidebar, click Assistants. You’ll see a list of all existing assistants for your account.
2

Click Create New Assistant

Click the Create New Assistant button in the top-right corner. The creation dialog opens.
3

Choose a template

Select the template that best matches your use case. Each template comes pre-loaded with a starting system prompt, a default voice, and an inbound or outbound direction.You can search templates by name, description, or category using the search box in the dialog.
4

Name your assistant

Type a name in the Assistant name field. The name pre-fills based on the template you selected, but you can change it to anything that helps you identify the assistant later (for example, “EU Support Bot” or “Q2 Outreach Caller”).
5

Click Create Assistant

Click Create Assistant. The assistant is created and you land on its detail page, where you can configure each tab.

​
Configure the Model tab

The Model tab controls which language model the assistant uses to generate responses and how it behaves.
1

Select a provider

Choose a provider from the Provider dropdown. Talkturo supports:
2

Select a model

After choosing a provider, the Model dropdown populates with the available models for that provider. The first model in the list is selected by default.
For most voice call use cases, a “mini” or “flash” variant gives the best balance of response speed and quality. Faster responses reduce the pause the caller experiences between speaking and hearing a reply.
3

Set Temperature

Use the Temperature slider (0–2) to control how varied the assistant’s responses are.
  • Lower values (0–0.5): More focused and consistent — good for support or compliance-heavy scenarios.
  • Higher values (1–2): More creative and varied — better for conversational or sales assistants.
4

Set Max Tokens

Enter a value in Max Tokens to limit how long each assistant response can be. The default placeholder is 250 tokens, which is a reasonable starting point for a voice call where responses should be concise.
Very high max token values can cause long pauses on calls because the assistant generates more text before speaking. Keep responses short for voice interactions.
5

Save your changes

Click Save Changes at the top of the assistant page to apply your model settings.

​
Configure the Transcriptor tab

The Transcriptor tab controls how the assistant converts the caller’s speech to text. Choosing the right provider and language directly affects transcription accuracy.
1

Select a provider

Choose one of the supported speech-to-text providers:
2

Select a model

After picking a provider, choose the model that fits your use case. Deepgram Nova-3 is a strong general-purpose choice. Use Nova-3 Medical or Nova-2 Medical for healthcare conversations with medical terminology. Use AssemblyAI Universal Streaming Multilingual when callers may switch between languages mid-call.
3

Select a language

Choose the primary language your callers speak. Available languages depend on the model selected. For example:
  • Deepgram Nova-3 and Nova-2 support English (US, UK), Spanish, French, German, Portuguese, and a Multilingual mode.
  • AssemblyAI Universal Streaming Multilingual adds Italian and Portuguese.
  • Cartesia Ink Whisper adds Chinese, Japanese, Korean, and Russian.
  • ElevenLabs Scribe V2 Realtime supports English, Spanish, French, German, Portuguese, Japanese, and Chinese.
4

Save your changes

Click Save Changes to apply the transcription settings.