Skip to main content
Build intelligent voice agents that understand, respond, and engage with your customers naturally. This guide walks you through every tab in the SquawkVoice Studio so you can launch a production-ready agent in minutes.

​
Overview

An AI Agent combines natural-language understanding, voice synthesis, and workflow automation. Each agent is fully configurable through four main tabs inside the Studio:
  1. Voice & Guidelines – Define how the agent greets callers, which voice it uses, and the system guidelines that steer its behaviour.
  2. Knowledge Base – Grant the agent access to company knowledge at runtime.
  3. Actions – Orchestrate API calls before, during, and after a conversation so the agent can fetch data, carry out tasks, and hand off results.
  4. Settings – Fine-tune transcription with custom keywords and other preferences.

​
1 · Voice & Guidelines Tab

Need to reshape variable values before you drop them into a prompt? Create a Data Transformer and pick it from the variable popper wherever you use {{variables}}.

​
Arabic Agents and Voices

Arabic is a fully supported agent language. In the voice catalogue set Provider to Munsit (Arabic) to see a large library of Arabic voices, each listed with its language, gender and dialect — Modern Standard Arabic (Fusha), Najdi, Hijazi, Emirati, Kuwaiti, Qatari, Bahraini or Omani. There is no plan gate and no beta flag. Arabic (AR) is also accepted as an agent’s Primary Language. That save was previously rejected, because the older transcription model had no Arabic at all.
These voices are not marked as Flash voices, and the catalogue sorts Flash voices first — so use the Provider filter or the search box rather than scrolling. They appear once the voice catalogue has synced for your region.
On a call, an Arabic agent now starts speaking as soon as its opening sentence is ready and long enough, rather than waiting for the whole reply to be written. Arabic question marks (؟) and semicolons (؛) are recognised as sentence endings; the Arabic comma (،) is not, because it separates clauses rather than sentences.

​
The Arabic Language Lock

These voices speak only Arabic, so selecting one converts the agent in a single step: single-language mode, Arabic as the primary language, and every other configured language dropped. Studio confirms with “Language switched to Arabic” and “Munsit voices only support Arabic. Translated messages will be ready in a few minutes.” The Languages card then shows itself pinned — the multilingual toggle is greyed out and inert, and the card explains “Locked to Arabic — the selected Munsit voice only supports Arabic. Choose a different voice to change the language.” The rule is enforced on the server, not just in the screen, so Squawk Copilot, an agent import and a direct API update all end up with a speakable configuration. An imported bundle that pairs an Arabic voice with another language is quietly corrected rather than rejected.
Test this on a duplicate before doing it to a live agent.
  • A multilingual agent that handled English, Spanish and French keeps only Arabic. The other languages are gone.
  • Your transfer, closing, silence and max-duration messages are re-translated into Arabic in the background, which takes a few minutes. Your greeting is not translated — the agent will still open the call in whatever language you wrote it in. Rewrite it yourself.
  • If the agent’s summary language was anything other than English, Arabic or Auto, it is reset to English.
  • The switch is one-way. Moving the agent back to another provider’s voice does not restore the languages you lost; it only releases the synthesis model back to Flash v2.5. Set the language yourself if you want the agent to stop being Arabic.
The translations finish in the background, and the agent’s page isn’t refreshed when they do. Greeting & Sign-off (the closing message), Handoff (the transfer message) and Agent Settings (the silence and max-duration messages) won’t show them until you switch agents or reload.

​
Arabic Accent

While an Arabic voice is selected, the Languages card gains an Accent picker, described as “Sets the regional variety the agent listens for and speaks in.” One choice tunes both halves of the conversation, so transcription and pronunciation can never disagree. Eighteen options are offered: Arabic — any region (the default), Modern Standard Arabic (Fusha), Emirati (ar-AE), Saudi (ar-SA), Qatari (ar-QA), Kuwaiti (ar-KW), Egyptian (ar-EG), Jordanian (ar-JO), Lebanese (ar-LB), Syrian (ar-SY), Palestinian (ar-PS), Iraqi (ar-IQ), Moroccan (ar-MA), Algerian (ar-DZ), Tunisian (ar-TN), Sudanese (ar-SD), Chadian (ar-TD) and Iranian Arabic (ar-IR).
  • Picking a regional option pins the transcription engine to that variety — noticeably fewer misheard words with strongly regional callers.
  • Arabic — any region and Fusha transcribe exactly as Arabic always did, so choosing either changes nothing about recognition.
  • On the speaking side the provider accepts a narrower set of hints, so Emirati and Fusha change delivery directly while the other regional choices take effect through better recognition.
The picker exists only while an Arabic voice is selected — an Arabic agent on another provider’s voice has no Accent control. The setting is cleared if the agent later leaves Arabic, so it cannot sit as stale configuration on an English agent.
Existing agents default to Arabic — any region, so nothing changes until you choose. Once you do, transcribed wording can shift slightly — re-test any prompt logic that matches on exact transcribed phrases.

​
Voice tuning for Arabic voices

Arabic voices respond to Stability and Speed but have no similarity concept, so the Similarity slider is hidden on the Tone + energy card and its description reads “Tweak stability and pace.” rather than “Tweak stability, similarity, and pace.” A stored similarity value is left in place and applies again if you move the agent back to a voice that supports it.
Known gap: the older agent editor’s Voice Fine-tuning drawer still shows the Similarity slider for an Arabic voice, where it has no effect.
If the Arabic provider is momentarily at capacity mid-call, the platform retries briefly and then skips just that one reply, keeping the caller on the line — the next turn gets a fresh attempt. Failures that can never produce audio end the call immediately rather than leaving a caller in silence, with the bot_error disposition so the call is identifiable in Call History.

​
2 · Knowledge Base Tab

Link zero or at most 2 Knowledge Groups that the agent can search during a call to answer user’s queries. The AI automatically retrieves relevant passages at runtime, enabling precise, up-to-date responses without hard-coding content in prompts.
An AI Agent can only be linked to at most two Knowledge Groups

​
3 · Actions Tab

Actions let the agent talk to your backend systems via HTTPS. They come in three flavours:
The three phases are now named Pre-conversation Actions, In-conversation Actions and Post-conversation Actions, because actions run on chat and email as well as on calls. This is a rename only — the phases, and every action inside them, behave exactly as before, and nothing needs reconfiguring.

​
3.1 · Pre-conversation Actions

  • Runs before the agent speaks.
  • Common use-case: FetchContact – pull CRM details using an API Request node and store them in variables like {{firstName}}, {{accountStatus}}.
  • Variables created here are available to personalize Initial Message and Guidelines.
  • Variables values are available throughout the lifetime of the interaction and can be utilized in In-conversation or Post-conversation Actions.
Pre-conversation actions are configured as visual flows. You can configure multiple Pre-conversation Actions which will be executed sequentially in the order they were created. This allows you to chain API calls and build up context before the conversation starts.
Only variables extracted from Pre-conversation Actions can be used to personalize the Initial Message and Guidelines. No other variable values are available at this stage.
See Pre-conversation Actions for the full walkthrough.

​
3.2 · In-conversation Actions

  • Triggered on demand when the caller asks for something (e.g. “CheckOrderStatus”).
  • Each execution is logged in the live transcript and action logs, giving you full visibility into what happened during the conversation.
  • Provide comprehensive documentation for each action, including specific trigger conditions and use cases. Detailed context helps the AI model accurately determine when and how to execute the action.
  • Define parameters to specify the data requirements for action execution. Each parameter can be configured with:
    • Name
    • Description (use this to provide format requirements and context to help the AI collect data correctly)
    • Type (String, Number and List)
    • Required/Optional flag Parameters are optional and you can configure zero or more as needed for the action.
  • You can define filler messages that the agent speaks while waiting for the API response using the Say node in the flow builder
  • External API calls can be executed through the flow’s API Request Node
In-conversation Action names must only contain letters, numbers, underscores or hyphens, and be between 1-63 characters long (e.g. use check_order_status or CheckOrderStatus, not Check Order Status!)
See In-conversation Actions for the flow builder in detail.

​
3.3 · Post-conversation Actions

Executed when the conversation ends—whether the caller hangs up, self-serves successfully, or escalates to a human. These are configured as flows that run asynchronously. Typical tasks include:
  • Generating a conversation summary and disposition
  • Saving the transcript to your CRM or data warehouse
  • Kicking off downstream workflows (e.g. ticket creation using flow nodes)
See Post-conversation Actions for the full walkthrough.

​
4 · Settings Tab

Use this tab to enhance speech-to-text accuracy and summary output:
  • Transcription Keywords – Add product names, acronyms, or industry-specific terms (e.g. medical jargon) that the transcription engine might otherwise miss.

​
Summary Language

Choose the language AI-generated call and chat summaries are written in. The setting sits in the Languages & localization card as Summary language, so you configure it alongside the languages the agent speaks and save both together.
  • Pick any of the agent’s configured languages, or Auto (detect from conversation) to match the language the caller actually used. Defaults to English.
  • The whole summary is written in that language — including the Additional Concerns section, which previously always came back in English.
  • Section headings (Issue, Resolution, Next Steps, Additional Concerns) stay in English; the narrative beneath each heading is written in the chosen language.

​
Greeting & Sign-off and Handoff

Greeting & Sign-off and Handoff, both below the main configuration tabs, are each organized as Voice | Chat | Email tabs, so the wording for each channel sits in one predictable place.
  • Greeting & Sign-off — the two tabs previously labelled Voice Greeting & Sign-off and Chat Greeting & Sign-off are now simply Voice and Chat, joined by a new Email tab.
  • Handoff — where this section previously stepped you through Outcome options and Transfer settings, it now presents the same three channel tabs.
Your existing voice and chat settings are unchanged — they have only moved under their own tab. Nothing needs re-entering after the upgrade.
For what the Email tabs hold, see Email Agents.

​
Phone Numbers on an Agent

An agent’s header carries an Assigned number chip — Assigned numbers when it holds more than one, listed most recently assigned first, showing the first two with +N for the rest and the full list on hover. An agent with no live assignment shows no chip. Numbers can be attached and detached from the agent itself: the ⋯ menu on an agent’s card in Build → AI Agents offers Assign phone number / Assign another number and Unassign phone numbers (N), both opening a multi-select drawer. See Phone Numbers. The numbers assigned to an agent are also the caller IDs its outbound voice campaigns call from, so assigning or unassigning one changes where those campaigns call from as well as which agent answers it. See Numbers as Campaign Caller IDs.

​
Variables Manager

The Build hub includes a dedicated Variables capability for managing every variable an agent uses in one place:
  • Create, edit, and delete custom variables with descriptions.
  • Filter for unused variables to keep the agent tidy.
  • Open View references on any variable to see exactly where it is used — across pre-conversation, in-conversation, and post-conversation actions (down to the flow node) and assistant settings.
  • Variables that are still referenced cannot be deleted, so flows never break silently.

​
Default values

A variable can carry a Default value — the value it starts every interaction with, before anything has had a chance to assign to it. Set it in Build → Variables, both in the drawer where you create a variable and in the edit drawer on any existing row; the field is optional, and its placeholder reads Optional starting value. The variables table shows a Default value column, after Name and Type, with either the stored value or a dash. Every interaction — voice, chat, or email — begins with that value already in place, so a {{variable}} used in Initial Message, in Guidelines, or in an action’s request body renders the default instead of an empty string when nothing has assigned to it yet.
The create drawer puts it plainly: “Every interaction starts with this value. Anything the agent assigns during the interaction replaces it. Leave empty to start unset.” The edit drawer reads: “Every interaction starts with this value. Clearing it makes the variable start unset again. Default values set on action assignments are not affected.”

​
Which value wins

​
Behaviour worth knowing

  • A default can be up to 2,000 characters. A blank or whitespace-only entry is stored as no default at all.
  • Every variable that existed before this release has no default, so nothing changes until you fill the field in.
  • The edit drawer always submits the field, so saving an edit with Default value empty clears a default you set earlier. Re-enter the value if you only meant to change something else on the row.
  • A variable created on the fly from an API Request node’s Response Mapping inherits that row’s Default Value (optional) as its own default.
  • A Decision node that tested a variable for emptiness may take a different branch once that variable has a default, because the variable is no longer empty at the start of the interaction. Review those conditions when you add a default to a variable a flow branches on.

​
Autosave & Draft Safety

  • Changes to the Name, Initial Message, Guidelines, voice selection/settings, handoff dispositions, transfer messaging, call end message, and advanced settings save automatically within a second or two after you pause typing.
  • A toast in the bottom-right (for example “Voice updated”) confirms each successful sync. If something fails you will see an error toast so you can retry.
  • If the Co-pilot changes the agent while you have an edit waiting to save on Agent Settings or Handoff, both changes are kept. If you both changed the same setting, yours is saved. The agent’s languages count as one setting: if you changed its languages or language mode, your whole language setup is kept.
  • An open Guidelines editor or Voice Configuration panel switches to the Co-pilot’s version as soon as it lands. The Co-pilot only sees what has been saved, so give a hand edit a moment to save before you ask it to build on that edit.

​
Live Testing Console & DTMF Keypad

1
Click Talk in the agent’s header to start a test call from your browser. Chat, beside it, opens a test chat instead.
2
Use the on-screen keypad to send digits (0–9, *, #) while the call is live.
3
Watch the transcript and duration panes to confirm the agent’s behaviour.
DTMF presses behave like caller speech: they barge-in to stop filler audio, appear in the transcript, and can trigger flows that expect keypad input.
The Co-pilot panel stays open while you test. Chat opens the test chat beside the panel, so you can try a change next to the conversation that asked for it. A Talk call covers the whole window until you end it, and your Co-pilot conversation is still there when you hang up.

​
Testing Transfers

When a Talk test call transfers to a phone number or a SIP address, the destination sees the agent’s own number as the caller ID. That’s the earliest-assigned of the agent’s numbers bought in the Studio: assigning more numbers later doesn’t change it, and unassigning it moves the caller ID to the next-earliest.
  • Numbers you brought yourself (BYO) are never used.
  • Numbers with an unresolved payment failure are never used either. When a subscription payment fails, every number you bought is flagged, so none of them qualifies until the payment is resolved. See When a Payment Fails.
  • The Assigned numbers chip in the agent’s header lists the most recently assigned number first, so the caller ID may not be the first number it shows.
If the agent has no qualifying number, the transfer may fail and the test call ends. Assign a number bought in the Studio to the agent before you test a transfer.
Transfers on real phone calls and SIP calls are unaffected by this, and that includes a campaign’s Test Call. It’s a real phone call rather than a browser test, placed from one of the agent’s numbers under the campaign caller-ID rules, so a transfer on it behaves as it would on any other real call.

​
Copying, Duplicating & Transferring Agents & Actions

​
Copy & Paste Actions

  • Open the Actions tab of an agent, locate the action you want to copy, click its actions menu (...), and choose Copy Action.
  • Navigate to the Actions tab of another agent (within the same account) and click Paste Action (or the clipboard icon) at the top of the action list to paste it. This copies the entire action configuration, parameters, and its linked flow.

​
Duplicate

  • Open the actions menu (...) on an agent card and choose Duplicate to create a Copy of {Agent Name} with the same actions, flows, variables, dispositions, linked knowledge groups, and data transformers.
  • Rename the clone and adjust it without touching the original agent.

​
Export

  • From the same menu select Export to download a JSON bundle that includes the agent, actions, flow definitions, knowledge links, dispositions, data transformers, and usage mappings.
  • Filenames include the export timestamp so you can keep versioned backups.

​
Import

  • Click Import above the table and upload a bundle that was exported from SquawkVoice Studio.
  • The importer recreates flows, reconnects knowledge groups (by name), restores data transformers (deduplicating when possible), and reports progress via toast notifications.

​
Best Practices

  • Start simple. Launch with a lightweight set of actions and expand iteratively.
  • Keep guidelines concise. The shorter the system prompt, the faster the agent responds.
  • Monitor analytics. Use the built-in dashboard to track resolution rates, average call duration, and customer sentiment.
  • Iterate on feedback. Real-world conversations reveal edge cases you can cover with new actions or guideline tweaks.

​
Next Steps

Ready to go deeper?

Chat Widget Integration

Embed the AI agent on your site and inject customer context dynamically.

Knowledge Base Management

Learn how to structure and maintain high-quality knowledge groups.

Call Analytics

Measure agent performance and identify optimisation opportunities.