AI Studio

Where your AI workforce becomes yours.

Name: AIWRKR AI Studio
Brand: AIWRKR

Create, evaluate and govern every asset your AI workers run on — voices, conversation flows, knowledge and integrations — powered by AQUAA, our sovereign AI model family, and deployed entirely inside your perimeter.

create/
  voice · model
  routing · knowledge
compose/
  workflow · integration
  packs
prove/
  assurance · registry

Registry · every asset, versioned

your tenant · nothing leaves the perimeter

Voice

customer-operations-hindi

v3

Promoted

Recorded on your floor, consent on file

Voice

voice-path-verification-2

v1

Candidate

7 samples · awaiting reference check

Flow

renewal-premium-due-reminder

v4

Promoted

Write behind a named approver gate

Flow

claim-status-enquiry

v2

Draft

Read-only end to end, no gate needed

Pack

insurance

v1.0.0

Imported

4 connectors · 9 operations · 4 flows

Knowledge

policy-servicing-wordings

v7

Promoted

Grace, revival, mode change, nominee

Model

aquaa-router-in

v2

Candidate

Scored against your own test set

The workbench

Three rooms, one tenant

Everything an AI worker runs on is made in one of three rooms. Create the assets, compose them into working processes, then prove them before a single customer hears the result.

Create

The assets

The raw material an AI worker speaks, thinks and reads with — each one made here and owned by you.

Voice

A voice built from a real person's recording, with consent stored beside it

Model

AQUAA™ models tuned on your own material, scored before promotion

Routing

Which worker takes which intent, and where a person enters

Knowledge

Your wordings, SOPs and policy language, versioned like code

Language

Additional language coverage beyond the 23 already shipped

SOON

Compose

The processes

Assets assembled into work: the flows a worker runs, the systems it touches, and the packs that arrive pre-wired.

Workflow

Flows built node by node, with gate classes on every action

Integration

Connectors bound to your CRM, core systems and ticketing

Packs

A working industry set — connectors, operations and flows — imported as drafts

Workforce

Compose whole teams of workers and their supervision

SOON

Prove

The evidence

Nothing reaches a customer on a feeling. Score it against your own tests, then promote what passes.

Assurance

Voices, flows and models scored against your golden test sets

Registry

Every asset, every version, and what changed between them

Flow sessions

Replay a run node by node and see where each gate fired

Lineage

Where an asset came from, and whether a pack update applies

Create · Voice

Your brand's voice. Actually yours.

Cloud AI voices are rented — the same synthetic persona, shared by everyone, billed per character. AI Studio lets you create a voice your company owns, three ways:

Every voice is built from a real person's recording, with documented consent — and it speaks for your brand in 23 languages, without ever leaving your infrastructure.

01

Record your own people

That warm, trusted voice on your service floor becomes the voice of your brand — recorded in the Studio, with consent captured and stored alongside the voice itself.

02

Bring a professional artist

Commission a studio recording from a voice artist. The recording, the rights and the resulting voice are yours by contract — forever.

03

Or hand it to us

The AIWRKR team produces your brand voice end-to-end — sourcing, recording, building and tuning — and hands you the finished, owned asset.

Create · Voice

Three steps, and a consent record that outlives the project

Record the person whose voice this is, attach the consent that permits it, and build.

1 Which voice

Pick the voice this recording belongs to, or create one. Name it for the role it plays, not for the person — the speaker is recorded separately.

· Existing voices show their version and sample count

· A new voice creates its first draft in one step

· Never built is stated plainly, not hidden

2 The reference

A connected passage of natural speech. Reading a paragraph aloud works better than repeating a short line.

· Record in the browser or upload a file

· Converted here to 16-bit PCM WAV

· Speaker name captured with the sample

3 Consent

The person in this recording has consented to their voice being cloned and used by this workforce. Nothing builds without it.

· An explicit consent confirmation, not a checkbox in a contract

· A consent note — the ticket, the signed form, the date

· Stored with the voice, for as long as the voice exists

Engines · probed live

AQUAA-Vox reachable

AQUAA-TTS-IN1 reachable

AQUAA-STT-HI1 reachable

Health only. Engine selection is not offered here — a build takes the server's default.

What the check looks for

At least 6 seconds of connected speech

More speech than silence

Under 2% of samples clipped

A room quiet enough to hear the voice above it

The Studio measures these when you save, and says which one failed.

Compose · Flow sessions

Watch a run before you trust it.

A flow session replays a run node by node: where it went, which gate fired, and what each tool call answered. Every step resolves from its declared fixture, so nothing calls a live system and nothing leaves the tenant.

Then ask the harder question — could this run live? The Studio answers it directly, and lists every reason it cannot.

Three run modes

Simulated Every tool step answers from its declared fixture. Nothing calls a customer system.

Sandbox Real connectors against your non-production estate, with writes still gated.

Live Only a promoted flow, and only once you have ratified the gate map.

Session ba74708e

Simulated Declared fixtures Succeeded

status_given · channel test

Graph

Lit where the session went. A node it never reached is drawn, not dimmed away — unreached is not the same fact as unavailable.

start

trigger.

inbound

→

identify

tool.call

→

claim

tool.call

→

tell

say

→

done

end

human

handoff · on error

Trace

#0 start trigger.inbound simulated 0 ms

#1 identify tool.call simulated 0 ms

#2 claim tool.call simulated 0 ms

#3 tell say simulated 0 ms

#4 done end simulated ended 0 ms

Inspect in mode

What would happen, without running anything. This is a read.

Simulated

Sandbox

Live

Every tool step answers from its declared fixture. Nothing leaves this system.

Tool steps

CRM.getCustomer

simulated · fixture for "getCustomer"

Claims.getClaimStatus

simulated · fixture for "getClaimStatus"

Could this run live?

No — and each reason is below.

· the flow version is draft — only a promoted version runs live

· the customer has not ratified the gate map — our classes are a proposal

Compose · Packs

Industry packs

A pack is a working set of connectors, operations and flows for one industry. Importing one replicates it into this tenant as drafts, with lineage recorded — your changes never flow back into the pack, and a pack update can be offered as a diff rather than applied.

Packs available to import — each arrives as drafts with simulated bindings only
Pack	Version	Connectors	Operations	Flows	What it covers
Banking and financial services	banking-v1 · v1.0.0	3	8	4	Card disputes, balances and statements for a retail bank
Insurance	insurance · v1.0.0	4	9	4	Renewal, claim status and policy servicing for a life insurer
Healthcare	healthcare-v1 · v1.0.0	4	9	4	Appointments, report status and billing estimates for a hospital group. No clinical advice
BPO and outsourced contact centres	bpo-v1 · v1.0.0	3	6	4	Collections callbacks, dispositions and campaign state for an operator running many end clients
Public sector and PSU	publicsector-v1 · v1.0.0	4	6	4	Citizen service enquiries, application status and grievance intake

Worked example · insurance · v1.0.0

What is actually inside a pack

Renewal, claim status and policy servicing for a life insurer. Every operation carries a gate class, and every flow ends somewhere you can point at.

Gate classes are a proposal. Live mode requires the customer to ratify them.

Connectors & operations

PolicyAdmin POLICY_ADMINISTRATION · SOAP

Fits: Guidewire / Duck Creek / Sapiens class

getPolicy READ acts alone

Policy status, premium, due date, sum assured

changePaymentMode WRITE needs authorisation

Change premium frequency. Material — needs a named approver.

updateNominee WRITE needs authorisation

Billing BILLING_COLLECTIONS · REST

getDueSchedule READ acts alone

sendPaymentLink WRITE needs authorisation

Sends a payment link to the registered number. A write because it contacts the customer, even though it moves no money.

refund DESTRUCTIVE never alone

Claims CLAIMS · REST

getClaimStatus READ acts alone

CRM CRM · REST

getCustomer READ acts alone

Identify the caller against the customer record

logInteraction WRITE needs authorisation

Write the interaction summary back to the customer record

Flows

Renewal — premium due reminder

The single flow that must complete end to end in a demo. It identifies the caller, reads the policy, states the amount and date, confirms the date back, and lands a write behind a gate.

start

trigger.inbound

→

identify

tool.call

→

policy

tool.call

→

state

say

Claim status enquiry

Read-only end to end. Demonstrates the acts-alone class without a gate.

start

trigger.inbound

→

identify

tool.call

→

claim

tool.call

→

tell

say

Renewal due — outbound reminder with payment link

Outbound is where persistency is won. Event-triggered: the due date, not a caller, starts this flow.

start

trigger.event

→

fetch

tool.call

→

state

say

→

confirm

extract

Premium mode change

A servicing write on the policy itself — the alteration class of work.

start

trigger.inbound

→

identify

tool.call

→

fetch

tool.call

→

state

say

KnowledgePolicy servicing wordings — Grace periods, revival, mode change and nominee rules.

Prove · Assurance & Registry

Proven before it ever meets a customer

Studio treats AI assets the way engineering treats software: nothing is promoted on a feeling.

Your own test sets

Score voices, flows and models against golden sets written from your own call history — not a vendor benchmark.

Promotion gates

Draft → candidate → production, with a recorded reason for every promotion and the evidence attached.

Versioned everything

Every voice, flow and knowledge base carries a version and a lineage. You can always see what changed, and roll back.

The promotion path

state: draft

Draft

Being built. Runs in simulated mode only, against declared fixtures.

state: candidate

Candidate

Complete and scored, awaiting the gate. Sandbox runs allowed, writes still held.

state: production

Production

Promoted with evidence on file. The only state that can answer a real customer.

What a gate checks before an asset is promoted
Asset	Scored against	Promoted when
Voice	Reference quality checks and your own listening panel	Speech, silence, clipping and room checks all pass
Flow	A golden set of runs, each with an expected outcome	Every run reaches its expected node, gates included
Model	Your test set for intent, entity and language accuracy	It beats the version currently in production
Knowledge	Retrieval accuracy against questions your floor actually gets	Answers cite the right wording, with no invented clauses