Assistant Configuration

Configure prompts, models, voice, telephony, tools, analysis, webhooks, and email summaries for Talkturo AI voice agents.

Configure every assistant tab

Assistant configuration controls how your voice agent speaks, listens, places calls, handles tools, and reports outcomes. Use this page as the field-level reference for all 10 configuration tabs in the assistant editor.

Some fields change based on the selected mode, provider, or enabled feature. When a field appears only in Standard mode or for a specific provider, that dependency is called out in the field description.

Prompt tab

The Prompt tab defines what the assistant says first and how it should behave for the rest of the call. Both prompt fields support CRM variables, so you can personalize calls with campaign, contact, and company data.

Prompt fields

first_messagestring

The first spoken sentence when a call connects. This field supports CRM variables such as {{contact.first_name}} and {{company.name}}.

system_promptstring

The full assistant instruction prompt. This field is shown as a monospace, expandable textarea and supports CRM variables.

prompt_templatestring

Optional starting template for the system prompt.

Allowed values:B2B SaaS Sales Qualify (Outbound)No Template

CRM variable picker

Press Ctrl + Space in a supported prompt field to open the variable picker and insert a variable at the cursor position.

Campaign variables

{{campaign.name}}
{{campaign.description}}
{{campaign.from_number}}

Contact variables

{{contact.first_name}}
{{contact.last_name}}
{{contact.full_name}}
{{contact.email}}
{{contact.phone}}
{{contact.job_title}}
{{contact.lead_status}}

Company variables

{{company.name}}
{{company.website}}
{{company.industry}}
{{company.phone}}
{{company.city}}
{{company.state}}
{{company.country}}

Custom field variables

{{contact.custom.<key>}}
{{company.custom.<key>}}

Use custom field variables when your CRM stores account-specific data such as renewal date, plan tier, territory, or owner name. Replace <key> with the custom field key from your CRM schema.

Prompt actions

Action	What it does
AI Improve	Sends the current prompt to `/api/improve-prompt` and returns an improved version
AI Translate	Translates the prompt to a supported language
Template selector	Replaces the current template choice with `B2B SaaS Sales Qualify (Outbound)` or `No Template`

Supported translation languages

AI Translate supports these 15 languages:

German
English
Spanish
French
Italian
Portuguese
Dutch
Polish
Russian
Turkish
Arabic
Hindi
Japanese
Chinese
Korean

Model tab

The Model tab controls the LLM provider, model, and generation settings. The available fields change depending on whether the assistant runs in Standard mode or Realtime mode.

Standard mode separates the language model, transcription, and text-to-speech pipeline. It exposes the widest provider set and the most tuning options.

modestring

Set to standard to use the non-realtime model pipeline.

Allowed values:standardrealtime

providerstring

LLM provider for Standard mode.

Allowed values:talkturoopenaigoogledeepseek-aiqwenmoonshotaianthropiccerebras

modelstring

Model name selected from the provider-specific dropdown. Available values depend on the selected provider.

temperaturenumber

Sampling temperature. Range 0 to 2, step 0.1, default 0.

max_tokensinteger

Maximum number of tokens the model can generate.

top_pnumber

Nucleus sampling setting for Cerebras only. Range 0 to 1, step 0.05, default 1.

reasoning_effortstring

Reasoning level for Cerebras thinking and gpt-oss models only.

Allowed values:defaultlowmediumhigh

Standard mode providers

Provider value	Display name
`talkturo`	Talkturo Fast
`openai`	OpenAI
`google`	Google
`deepseek-ai`	DeepSeek AI
`qwen`	Qwen
`moonshotai`	Moonshot AI
`anthropic`	Anthropic
`cerebras`	Cerebras

Realtime mode combines speaking and model response into a low-latency streaming flow. It supports fewer providers but is optimized for live conversations.

modestring

Set to realtime to use a streaming realtime model.

Allowed values:standardrealtime

providerstring

Realtime model provider.

Allowed values:openaigoogle

modelstring

Realtime model name. OpenAI uses gpt-realtime or gpt-realtime-mini. Google uses Gemini Live models.

voicestring

Built-in realtime voice used by the provider.

temperaturenumber

Sampling temperature. OpenAI supports 0.6 to 1.2. Google supports 0 to 2.

Realtime providers and voice options

Provider value	Display name	Example models	Voice options
`openai`	OpenAI Realtime	`gpt-realtime`, `gpt-realtime-mini`	`marin`, `cedar`, `alloy`, `ash`, `ballad`, `coral`, `echo`, `sage`, `shimmer`, `verse`
`google`	Gemini Live	Gemini realtime models	`Puck`, `Charon`, `Kore`, and other Google voice options

Voice configuration behaves differently by mode. In Realtime mode, voice selection happens inside the Model tab. In Standard mode, use the separate Voice tab.

Voice tab

The Voice tab is available in Standard mode only. It controls TTS provider selection, voice identity, speaking style, playback speed, and synthesized background sound.

This tab does not apply to Realtime mode. Realtime voices are configured in the Model tab because the provider handles generation and speech together.

Voice fields

voice_providerstring

Text-to-speech provider.

Allowed values:cartesiaelevenlabsinworldopenaidashscopekugel

voice_modelstring

Provider-specific voice model. The list is loaded dynamically from /api/voice-providers/models.

voice_idstring

Specific voice selection for the chosen provider and model. The list is loaded dynamically from /api/tts-voices.

emotionstring

Speaking emotion for Cartesia only.

Allowed values:neutralexcitedsadangry

background_ambiencestring

Background ambience mixed into synthesized speech.

Allowed values:noneoffice_ambiencekeyboard_typingkeyboard_typing2hold_music

ambience_volumenumber

Ambience volume. Range 0 to 1, step 0.05, default 0.8.

thinking_soundstring

Sound effect played while the assistant is thinking.

Allowed values:nonekeyboard_typingkeyboard_typing2keyboard_typing_mix

noise_cancellationstring

Noise cancellation mode. bvc is the default.

Allowed values:offncbvc

speednumber

Speech speed multiplier. Range 0.5 to 2.0, step 0.1, default 1.

Noise cancellation modes

Value	Meaning	Default
`off`	No noise cancellation	No
`nc`	Standard noise cancellation	No
`bvc`	Background Voice Cancellation	Yes

Transcriptor tab

The Transcriptor tab is available in Standard mode only. It controls the speech-to-text provider, recognition model, language, and provider-specific streaming settings.

Transcriptor fields

providerstring

Speech-to-text provider.

Allowed values:deepgramassemblyaicartesiaelevenlabsinworld

modelstring

Provider-specific transcription model, such as deepgram/nova-3, deepgram/nova-2, or assemblyai/universal-streaming.

languagestring

Language code used for recognition. Supported values depend on the selected model and can include codes such as en, en-US, de, es, fr, it, pt, and multi.

keytermsarray[string]

Custom terms to improve recognition accuracy. You can add up to 100 keyterms.

Advanced transcription settings

Add company names, product names, competitor names, and industry terms to keyterms before testing a new campaign. That usually improves transcript quality more than changing the base model alone.

Telephony tab

The Telephony tab controls phone number assignment, calling permissions, test calls, compliance playback, and voicemail behavior.

Telephony fields

assigned_phone_numbersarray[string]

Phone numbers assigned to the assistant. This is a multi-select sourced from the workspace phone number inventory.

region_or_sip_trunkstring

Region or SIP trunk selection for Telnyx accounts.

inbound_calling_enabledboolean

Enable inbound calling for this assistant.

inbound_companystring

Company selected for inbound calling when inbound calling is enabled.

outbound_calling_enabledboolean

Enable outbound calling for this assistant.

destination_countriesarray[string]

Allowed destination countries for outbound calling.

test_call_from_numberstring

Source number used for test calls.

test_call_to_numberstring

Destination number used for test calls.

Compliance briefing

compliance_briefing_enabledboolean

Enable a compliance briefing at the start of the call.

compliance_templatestring

Predefined compliance briefing template.

compliance_custom_textstring

Custom compliance briefing text. Maximum 2000 characters.

require_acknowledgementboolean

Require the callee to acknowledge the compliance briefing before the assistant continues.

fallback_policystring

Fallback behavior if acknowledgement fails or the flow cannot continue as expected.

Allowed values:continueretryblock

Voicemail detection

voicemail_detection_enabledboolean

Enable voicemail detection. Default true.

voicemail_actionstring

Action taken when voicemail is detected.

Allowed values:leave_messagehangup

voicemail_messagestring

Voicemail message to leave when voicemail_action is leave_message. Maximum 2000 characters.

Review compliance and voicemail text before enabling outbound campaigns. These fields affect live call behavior and may have legal or policy implications for your use case and region.

Functions tab

The Functions tab attaches callable tools to the assistant. These tools let the assistant end calls, transfer callers, schedule meetings, trigger automations, or call external systems.

Built-in call functions

end_call.enabledboolean

Enable the End Call function.

end_call.trigger_phrasesarray[string]

Phrases that allow or encourage the assistant to end the call.

end_call.closing_phrase_modestring

How the assistant chooses its closing phrase before disconnecting.

Allowed values:fixednoneai

end_call.closing_phrasestring

Closing phrase used when closing_phrase_mode is fixed.

call_transfer.enabledboolean

Enable call transfer.

call_transfer.play_dialtoneboolean

Play a dialtone during transfer.

Transfer department fields

Each transfer department defines one destination the assistant can route to.

departments[].namestring

Display name for the department.

departments[].transfer_modestring

Transfer mode for the department.

Allowed values:warmcold

departments[].phone_numberstring

Phone number used for the transfer destination.

Connected tool integrations

cal.comobject

Cal.com integration. Supported functions are schedule, get_bookings, cancel, and check_availability.

zapierobject

Zapier integration. Each assistant supports a single Zapier instance.

custom_webhooksarray[object]

Custom webhook tools. An assistant can have multiple webhook tool configurations.

turboflow_workflowsarray[object]

Published TurboFlow workflows attached as callable tools.

Function availability summary

Function	Multiplicity	Notes
End Call	One configuration	Built in
Call Transfer	One configuration with multiple departments	Built in
Cal.com	One integration	Exposes scheduling actions
Zapier	One integration per assistant	Single instance only
Custom Webhook	Multiple	Each webhook has its own config
TurboFlow Workflows	Multiple	Attach published flows

Knowledge Base tab

The Knowledge Base tab connects processed documents from your workspace knowledge base to the assistant. Attached files become available for retrieval during calls.

Knowledge base fields

attached_documentsarray[string]

Documents attached to the assistant from the workspace knowledge base.

file_statusstring

Processing status of each file. Files must be processed before the assistant can use them.

chunk_countinteger

Number of chunks generated from the file.

character_countinteger

Number of characters in the file.

A file is usable only after it has been processed into chunks and embeddings. If a file is attached but still processing, the assistant will not retrieve from it yet.

Attached file list

Field shown in UI	Meaning
Status	Whether the file is ready for use
Chunk count	Number of indexed chunks
Character count	Source text size
Attachment state	Whether the document is currently attached to the assistant

Analysis tab

The Analysis tab defines what Talkturo should extract and evaluate after each call. Use it to capture structured fields, classify outcomes, score leads, and determine whether the call succeeded.

Analysis fields

custom_extraction_promptsarray[string]

Custom prompts for extracting structured post-call data.

outcome_classificationobject

Configuration for post-call outcome classification.

lead_scoringobject

Configuration for lead scoring logic.

success_evaluationobject

Configuration for determining whether a call counts as successful.

Webhooks tab

The Webhooks tab sends call lifecycle events to your systems. Configure the destination URL, choose which events to send, and use the built-in test flow before turning a webhook on for production traffic.

Webhook fields

webhook_urlstring

Destination URL that receives webhook events.

secretstring

Secret used to sign or validate webhook delivery.

eventsarray[string]

Selected event types for this webhook configuration.

Allowed values:call.startedcall.endedcall.transcribed

test_webhookaction

Sends a test webhook using the current configuration.

Supported events

Event	Trigger
`call.started`	A call begins
`call.ended`	A call finishes
`call.transcribed`	Transcription data becomes available

Run a test webhook after changing the URL, secret, or event set. That catches signing mistakes and endpoint mismatches before live calls start sending traffic.

Email Summary tab

The Email Summary tab sends an automatic call summary after a call completes. Use it when a human operator, manager, or shared inbox needs a copy of each completed call summary.

Email summary fields

enabledboolean

Enable or disable automatic email summaries.

recipient_emailstring

Recipient email address for the summary.

languagestring

Language used for the email summary.

auto_sendboolean

Summaries are sent automatically after call completion when enabled.

Mode-dependent tabs

Not every tab is available in every assistant mode. Use this matrix when you need to know where a setting lives.

Tab	Standard mode	Realtime mode
Prompt	Yes	Yes
Model	Yes	Yes
Voice	Yes	No
Transcriptor	Yes	No
Telephony	Yes	Yes
Functions	Yes	Yes
Knowledge Base	Yes	Yes
Analysis	Yes	Yes
Webhooks	Yes	Yes
Email Summary	Yes	Yes

What to configure first

Choose the right model mode

Standard mode gives you separate control over model, voice, and transcription. Realtime mode reduces latency and moves voice selection into the Model tab.

Set up telephony before testing

Assign numbers, enable inbound or outbound calling, and confirm voicemail and compliance behavior before placing live calls.

Was this page helpful?