VoiceAgent Expert Mode

This article explains how to see and edit your VoiceAgent’s full configuration in the Advanced section, including settings you can’t change anywhere else.

User level: Admin

Expert Mode gives you your VoiceAgent’s full configuration as JSON. It shows every setting, including the ones you’ve never changed, with their current values. You can fine-tune settings that aren’t available in the other sections, such as the AI model, voice behavior, and call handling.

Open Expert Mode

In the Dashboard, go to AI Voice Agents > Agents and open your VoiceAgent.
In the left menu, under Build, click Advanced.

The Advanced section of a VoiceAgent, with the Configuration panel showing the full JSON and the No issues status bar

The Configuration panel shows the full configuration. Above it, you can also ask the AI Composer to make changes for you, in plain words.

What you see

Every setting, with its value. Settings you haven’t changed show their default value, so you can see everything you can tune. You don’t need to save first to see them.
Settings for every speech provider. elevenLabsSettings and deepgramSettings are both listed, whichever provider your VoiceAgent uses.
A status bar above the JSON. It says No issues when the configuration is valid. If something is wrong, for example an unsupported value, it tells you what to fix before you can save.

Edit the configuration

Change the values you want in the Configuration panel.
Check that the status bar says No issues.
Click Save.

To undo your unsaved changes, click Revert. The other sections (Identity, Knowledge Base, and Skills & Scenarios) and the Advanced section show the same configuration in two different ways. A change you make in one is reflected in the other.

Copy a configuration to another VoiceAgent

In the Advanced section, click Copy to copy the full configuration, including the default values.
Open the other VoiceAgent’s Advanced section, replace its configuration with the copied one, and click Save.

Because the copy includes every setting, the other VoiceAgent behaves exactly the same.

Change the agentName. Each VoiceAgent needs a unique name.
Reconnect account-specific resources. A configuration can refer to things that belong to your account, such as knowledge bases, custom tools, phone numbers, or a calendar integration. When you copy a configuration to another account, set these up again there.
After a successful import, save the VoiceAgent twice. A known set of edge cases only appears after the second save.

Settings reference

Identity

JSON field	What it sets
`agentName`	The VoiceAgent’s name. Must be unique in your account.
`direction`	Whether the agent handles inbound or outbound calls: `"inbound"` or `"outbound"`.
`language`	The language the VoiceAgent speaks.
`secondaryLanguages`	Additional languages the VoiceAgent can detect and switch to during a call. Optional.
`voice`	The voice used by the VoiceAgent.
`greeting`	The exact opening line the VoiceAgent says at the start of a call. If empty, inbound agents use a generic default; outbound agents open with silence.
`startSpeakingFirst`	Whether the VoiceAgent speaks first when the call connects (`true`), or waits for the caller (`false`).
`defaultOutboundNumberId`	The phone number the VoiceAgent calls from.
`failoverOutboundNumberId`	A backup number to call from, if the default number can’t be used.

Knowledge and behavior

JSON field	What it sets
`knowledgeBaseIds`	The knowledge bases the VoiceAgent can draw on when answering questions.
`goalPrompt`	The core instructions that define the VoiceAgent’s behavior during a call.
`skills`	Structured capabilities, for example `takeMessage` and `extractData`. Each skill has `enabled` and its own settings.
`scenarios`	How the VoiceAgent responds to specific situations. Each scenario has `when`, `reply`, and optionally an `action` such as `"hangup"`.
`guardrails`	Fixed responses for situations that need a consistent answer regardless of context.

In skills.extractData.properties, each property needs a unique name. If two properties share a name, the configuration isn’t saved.

Conversation style

These two fields control how the VoiceAgent speaks.

Field	What it sets	Values
`tone`	How the VoiceAgent sounds. A list of style presets; they can be combined, but one or two work best (each additional preset dilutes the others). Order doesn’t matter, duplicates are removed automatically, and free-text values aren’t accepted.	`"professional"`, `"friendly"`, `"casual"`, `"calm"`, `"empathetic"`, `"direct"`. Default `["friendly", "direct"]`.
`verbosity`	How much the VoiceAgent says per reply. `"concise"` = the shortest useful answer. `"standard"` = one to two sentences, offering more if needed. `"thorough"` = fuller replies with helpful context (still brief; this is a phone call).	`"concise"` / `"standard"` / `"thorough"`. Default `"standard"`.

Tone preset guide:

Preset	Effect
`professional`	Measured, precise, and respectful
`friendly`	Warm and personable
`casual`	Relaxed and informal, like a helpful colleague
`calm`	Steady, unhurried, and patient
`empathetic`	Acknowledges the caller’s feelings before moving to solutions
`direct`	Leads with the answer and keeps pleasantries to a minimum

Regardless of tone and verbosity, the VoiceAgent always stays brief during operational moments: confirmations, read-backs, farewells, and transfer or booking steps.

AI model

Field	What it sets
`llm`	The AI model the VoiceAgent uses. See the list of supported models below.
`provider`	The speech provider: `"elevenlabs"` or `"deepgram"`. In most cases, the language determines the provider automatically.
`temperature`	How varied the VoiceAgent’s responses are, from `0.0` to `1.0`. For professional or structured agents, `0.1` to `0.3` is recommended.

Supported models for llm

Model	Value
GPT-4o Mini	`openai - gpt-4o-mini`
GPT-4o	`openai - gpt-4o`
GPT-4.1	`openai - gpt-4.1`
GPT-4.1 Mini	`openai - gpt-4.1-mini`
GPT-4.1 Nano	`openai - gpt-4.1-nano`
GPT-5	`openai - gpt-5`
GPT-5 Mini	`openai - gpt-5-mini`
GPT-5 Nano	`openai - gpt-5-nano`
GPT-5.1	`openai - gpt-5.1`
Claude 3 Haiku	`anthropic - claude-3-haiku`
Claude Haiku 4.5	`anthropic - claude-haiku-4-5`
Claude 3.7 Sonnet	`anthropic - claude-3-7-sonnet`
Gemini 2.0 Flash Lite	`google - gemini-2.0-flash-lite`
Gemini 1.5 Pro	`google - gemini-1.5-pro`

Voice tuning

Field	What it sets	Values
`stability`	How consistent the VoiceAgent’s voice sounds throughout the call. Higher values produce a more stable voice.	`0.0`–`1.0`
`similarity`	How closely the voice matches the original voice model.	`0.0`–`1.0`
`optimizeStreamingLatency`	Trades voice quality for lower response latency. Higher values reduce latency but may affect quality. Don’t set it to `4`: it turns off text normalization, so numbers and dates are mispronounced.	`0`–`4`

Call behavior

Field	What it sets	Values
`maxCallDuration`	A hard limit on call length, in minutes.	Recommended: 3–5 for outbound, 15–30 for inbound.
`dialTime`	How long the VoiceAgent waits for an answer on outbound calls before giving up.	`5`–`120` seconds.

ElevenLabs settings

These settings apply when your VoiceAgent uses ElevenLabs, the default for most languages. They’re inside the elevenLabsSettings object, not at the root of the configuration.

Field	What it sets	Values
`speculativeTurn`	The VoiceAgent starts composing its reply before the caller has finished speaking, which reduces response time.	`true` / `false`
`turnEagerness`	How readily the VoiceAgent takes its turn when the caller pauses.	`"patient"` / `"normal"` / `"eager"`
`turnModel`	The version of the turn-detection model.	`"turn_v2"` / `"turn_v3"`
`vad`	Filters out background noise, so it doesn’t trigger an unintended response.	`true` / `false`
`asrKeywords`	Words or phrases to prioritize in speech recognition, such as product names or technical terms.	List of strings
`interruptionIgnoreTerms`	Short caller acknowledgements (for example, agreement sounds) that shouldn’t count as an interruption, so the VoiceAgent keeps talking through them. Written in the caller’s language.	List of strings
`interruptionIgnoreTermLanguages`	The languages the ignore terms are matched in.	List of language codes
`backgroundSound`	An ambient sound under the VoiceAgent’s speech. `null` means none.	`null`, or an object: `preset` (one of `"office"`, `"office2"`, `"restaurant"`, `"city"`, `"typing"`, `"elevator"`, `"elevator2"`, `"elevator3"`, `"elevator4"`), `volume` (`0.0`–`1.0`, optional), `crossfadeLoop` (`true`/`false`, optional; loops the sound seamlessly)

Example:

"elevenLabsSettings": {
  "speculativeTurn": true,
  "turnEagerness": "patient",
  "turnModel": "turn_v3",
  "vad": true,
  "asrKeywords": ["Capil Clinic", "injerto capilar", "Barcelona"],
  "interruptionIgnoreTerms": ["vale", "sí", "ajá"],
  "interruptionIgnoreTermLanguages": ["es"],
  "backgroundSound": {
    "preset": "office",
    "volume": 0.3,
    "crossfadeLoop": true
  }
}

When you save, the elevenLabsSettings object you send replaces the stored one. To clear a single setting, remove it from the object.

Deepgram settings

deepgramSettings holds the settings for VoiceAgents that use Deepgram. It’s empty when there’s nothing to set.

Managed by CloudTalk

minorVersion and hashes are managed by CloudTalk. Leave them as they are.

Need help or have a question? Just reach out through our Support portal. We’re here for you.

Get started

Numbers & porting

AI

AI Sales Dialer

CloudTalk Phone & Go

Messaging

Dashboard

Integrations

Billing & subscription

Tech News & Updates

Legacy dashboard

Open Expert Mode

What you see

Edit the configuration

Copy a configuration to another VoiceAgent

Settings reference

Identity

Knowledge and behavior

Conversation style

AI model

Voice tuning

Call behavior

ElevenLabs settings

Deepgram settings

Managed by CloudTalk