Take 10 minutes.
Get hours back every week.
Short walkthroughs that turn the platform into your operations team: automated pipelines, instant follow-ups, AI-handled admin. One video, one less thing to do manually.
Voice Agent Settings Reference
A reference for every setting on the Voice Agent detail page — model, voice, Flex Mode, data storage, fallback voices, and more.
The voice agent detail page has three tabs: Overview, Settings, and Channels. This article covers all eight fields in the Settings tab and explains what the Channels tab shows.
To open the detail page for any agent, go to https://app.trustpager.com/auto/voice-agents and click the agent name.
Settings tab
LLM Model
The language model that powers the agent's reasoning during a call. Two options are available:
- Claude 4.5 Haiku — fast and cost-efficient. The right choice for most agents: greetings, bookings, FAQ handling, lead qualification. Response latency is low, which makes conversations feel natural.
- GPT-4o — slower and more expensive, but better for flows that require multi-step reasoning, long-form instruction following, or complex branching. Only switch to this if Haiku is producing unreliable outputs in a specific flow.
TrustPager sets Claude 4.5 Haiku as the recommended default. Changing the model takes effect on the next published version — a draft save alone won't apply it to live calls.
Flex Mode
A flow-level toggle that controls how strictly the agent follows the node sequence you built.
- Off (default) — the agent follows the flow nodes in order. It moves from one node to the next only when an edge condition is met. Predictable and easier to debug.
- On — the agent can move between nodes more freely based on conversational context, without waiting for a strict edge match. Useful when conversations don't follow a predictable sequence and you want the agent to respond naturally to detours.
Start with Flex Mode off while building and testing. Turn it on only if callers are frequently going off-script in ways your edge conditions don't cover.
Data Storage
Controls what call data is retained after each call. Three modes:
- Store everything — full transcript, audio recording, and metadata are saved. Use this when you need complete records for QA, training, or compliance.
- Mask PII — the transcript and metadata are saved, but personal identifiers (names, phone numbers, email addresses, dates of birth) are redacted before storage. Use this when you want audit history without storing raw personal data.
- Store nothing — no transcript, no recording, no metadata beyond the call record itself. Use this for calls where privacy is paramount and you do not need a review trail.
TrustPager applies the recommended default of Store everything so you can review early calls and improve the flow. Change it once the agent is stable if your privacy obligations require it.
Voice
The voice the agent uses for all speech. The picker shows a library of available voices — each listing includes the language, accent, and a play button so you can preview it before committing. Click the play button next to any voice to hear a sample. Select a voice and save — it takes effect on the next published version.
The platform defaults to Australian English (en-AU). Match the voice to your caller base: an agent handling Australian inbound leads should use an en-AU voice; an agent handling US leads should use an en-US voice.
Fallback Voices
One or more backup voices the agent uses if the primary voice becomes temporarily unavailable. The agent switches to the first fallback automatically — callers experience a brief voice change but the call continues uninterrupted.
Add at least one fallback voice if your agent handles high call volumes or is critical to your operation. The same preview play button is available in the fallback picker.
Voice Model
The underlying synthesis engine used to generate speech from text. In most cases the default is correct — it is matched to the voice you selected. Only change this if you are experiencing audio quality issues and have been advised to try a different model by support.
Language
Sets the language for the agent's speech recognition (what it hears) and synthesis (what it says). Defaults to en-AU. Change this if your callers speak a different language — setting the correct language improves recognition accuracy significantly.
Normalize for Speech
When enabled, TrustPager pre-processes the agent's text output to make it more natural when spoken — for example, expanding abbreviations or formatting numbers. Leave this off in almost every case: if you have written your flow content to be naturally speakable (which you should), normalisation will add little and may introduce unexpected rewrites. Only enable it if you are feeding raw data (product codes, formatted dates, currency strings) directly into speech output.
Channels tab
The Channels tab gives you a unified view of every phone number and web widget wired to this agent. It does not support editing — it is a read-only summary. To manage any resource shown here, click the external-link icon on its row and you will be taken to its dedicated settings page.
The tab is divided into three sections:
- Inbound Numbers — phone numbers that route inbound calls to this agent. Click any row to open that number's detail page at https://app.trustpager.com/admin/phone-numbers. The Add link in the section header also navigates there.
- Outbound Calling — phone numbers configured as this agent's outbound caller ID. Managed from the same admin page at https://app.trustpager.com/admin/phone-numbers.
- Web Widgets — TrustPager websites that embed a call widget powered by this agent. Click any row to open the website's settings page. The Add link navigates to https://app.trustpager.com/admin/websites where you can wire up a new site.
If a section is empty, nothing is currently wired to this agent in that channel. Use the Add link or the dedicated admin page to connect a resource.
Recommended settings
TrustPager ships a recommended profile for new agents. The recommended defaults are:
- LLM Model: Claude 4.5 Haiku
- Flex Mode: Off
- Data Storage: Store everything
- Normalize for Speech: Off
These defaults work well for the majority of use cases. Adjust them once you understand how your specific flow behaves in live calls.
Two ways to
meet FinalPiece.
Talk to Evie right now, or book a real human for a deeper walkthrough. Whichever feels right.
Pick up the phone.
Hear her in 10 seconds.
- RealHer voice passes for human.
- SmartKnows your prices, hours, and trade.
- PatientPolite, attentive, never has a bad day.
Pencil us in.
60 minutes with a real human · Google Meet · Free.