statuslive · production pilot day167 aht-41% script compliance100% seats20 active residencydual-region in-house llmq4 2026 statuslive · production pilot day167 aht-41% script compliance100% seats20 active residencydual-region in-house llmq4 2026
AROOGA COPILOT · ACTIVE · v3.2.1

Your operators respond.
They don’t search.

Real-time agent assist with deterministic script execution. Sovereign by deployment today, sovereign by model tomorrow. Deployed in days, not months. Amplifies operators — doesn’t replace them.

uptime
99.9%
167 days since last incident
aht reduction
-41%
pilot data · medium confidence
onboarding
3-5 days
down from 2-4 weeks
in-house llm
Q4 2026
committed roadmap · on track
The Honest Frame

Sovereign by deployment today. Sovereign by model tomorrow.

Today · shipping now [ ACTIVE ]

Your data, your region. Multi-provider LLM orchestration.

On-prem or dual-region deployment. Self-hosted TTS. Deterministic script engine for routine flows. Multi-provider LLM abstraction — you choose which vendors are enabled per tenant, at the config layer.

on-prem k8s dual-region self-hosted tts deterministic scripts pii masking
Q4 2026 · committed roadmap [ ON TRACK ]

Your model. In-house. On your GPUs.

In-house distilled LLM ships as the default provider in the reference deployment. External providers remain available if you choose to keep them; the migration path is yours, not ours.

in-house llm provider-optional vertical-tuned committed roadmap sovereign by model
[ CASE REPORTS ]

>_ Three things your operators stop doing manually.

Every use case is live in a production pilot before it ships. No demo-only features.

Case 01 · Faster onboarding

Day-one operators perform like week-three operators.

Copilot suggests the right next question, the right follow-up, the right disclaimer — in real time, from your knowledge base. New hires follow the live suggestion pane instead of memorising 30-page scripts.

Onboarding 2-4 weeks → 3-5 days · pilot data, medium confidence
Case 02 · Script automation

Routine flows run deterministically. No LLM in call control.

Password resets, appointment confirmations, KYC checks, SLA lookups — your top call reasons follow a scripted path with deterministic execution. Auditable. No hallucination. LLM reserved for the 40% that needs judgement.

Scripts are state machines · LLM suggests phrasing only
Case 03 · Language expansion

Live STT and suggestion localisation.

Live speech-to-text plus suggestion localisation lets your ops team expand into new languages without hiring native speakers first. Language packs ship on a 2026 roadmap, with each new language validated before general availability.

Pilot language live · English shipping · TR + AR on 2026 roadmap
[ LIVE SESSION LOG ]

> what one assisted call looks like on the wire.

Every call, every suggestion, every script step is logged with timestamp, provider, region, and operator ID. Full export for compliance review, redacted for sample below.

session_7a4-f3 · redacted for public display streaming
14:02:17.044  CALL_START    operator=op-082 queue=billing region=eu-west
14:02:17.118  STT_CONNECT   provider=self-hosted · latency=74ms ok
14:02:22.851  INTENT        classified=refund_request confidence=0.96
14:02:22.874  SCRIPT_STATE  state=verify_identity · det=true
14:02:31.203  PII_MASK    tokens_masked=2 (card_last4, customer_dob)
14:02:31.311  LLM_SUGGEST   provider=llm-9b-tenant region=eu-west
                          prompt_tokens=1144 · latency=287ms
14:02:31.312  SUGGEST       "Can I confirm the last four digits of the card on file?"
14:02:47.601  SCRIPT_STATE  state=verify_identity → approve_refund
14:02:47.612  POLICY_CHECK  rule=refund_threshold · amount=48.00 pass
14:03:02.498  CRM_WRITE     provider=zendesk · ticket=#24981 ok
14:03:14.807  CALL_END      duration=57.8s · aht_delta=-38%
14:03:14.810  AUDIT_WRITE   rows_inserted=14 ok
[ PIPELINE ]

> listen. understand. assist. audit.

Four stages. Every stage instrumented. Every suggestion logged. Every LLM call tagged with provider and region.

t+0 msstage 01

Listen

SIP or WebRTC softphone. Sovereign telephony backbone. Live STT with sub-500 ms latency, VAD filtering, speaker diarisation.

SIP · WebRTC · STT < 500ms
t+230 msstage 02

Understand

Multi-provider LLM abstraction today, in-house distilled model Q4 2026. GraphRAG layer resolves context from your knowledge base — optionally backed by AROOGA CORTEX.

multi-provider · graphrag
t+420 msstage 03

Assist

Suggestion pane to the operator. Deterministic script execution where the flow is routine. LLM suggests phrasing; the state machine controls the call.

llm in copy · state machine in control
t+endstage 04

Audit

Every call, every suggestion, every script step logged with timestamp, provider, region, operator ID. Full export for compliance review.

per-call audit · per-period export
[ ARCHITECTURE ]

> three tiers. your dc. your keys. your provider policy.

Operator portal on top. Orchestration fabric in the middle. Sovereign platform at the bottom. Audit any call at the boundary between tiers.

tier layer components surface
L3 Application operator portal · admin console · script authoring · real-time transcript & suggestion pane web · desktop softphone
L2 Orchestration deterministic script engine · llm abstraction · self-hosted tts · graphrag multi-provider · auditable
L1 Sovereign platform sip telephony · dual-region stt · on-prem k8s · pii masking proxy your dc · your keys
[ COMPLIANCE ]

>_ Region, residency, audit — per tenant.

Your compliance team configures data residency during onboarding. Regions are enforced at the infrastructure layer, not a preference flag.

Dual-region data residency shipped Tenant-scoped region enforcement. More partner regions (Dubai, Istanbul, Virginia) in 2026.
On-prem K8s appliance shipped hostNetwork deploy. Runs inside your perimeter. No cloud dependency in the hot path.
Full audit trail per call shipped Timestamp, provider, region, operator ID per event. Customer-defined retention & export format.
PII masking shipped Integrated by default. Sensitive tokens masked at the operator edge before transmission.
No training on your data shipped Customer call data never enters model training sets. Architectural, not contractual.
SOC 2 on track Track opens 2026. Security questionnaire available under NDA.
GDPR / HIPAA-ready pattern pattern ready Architecture-ready. Market-by-market certification tracks announced on close.
[ MARKET POSITION ]

> what we ship that cresta, cognigy, observe.ai don’t.

Feature comparison based on public product pages as of April 2026. We audit quarterly. These are excellent cloud products — for a different buyer than you.

capability COPILOT cresta cognigy observe forethought
On-prem / air-gap deploymentcloudcloudcloudcloud
Multi-provider LLM (no single-vendor lock)coupledcoupledcloud LLMscoupled
Deterministic script enginepartial
Self-hosted TTS (no per-word cost)
Dual-region residency out of boxUS-onlyEU + USUS-primaryUS
Days-to-deploy (not months)daysweeksweeksweeksweeks
Published in-house model roadmapQ4 2026
[ PRODUCTION STATUS ]

> live in production since late 2025.

Real operators, real calls, published quarterly. No cover-photo renders.

LIVE · pilot day 167

20 operators · Eastern European contact-centre pilot · Sufler mode.

status
live
mode
Sufler assist
seats
20 operators
reporting
quarterly

Start with one queue. Run for four weeks. Keep the stack you have.

Paid pilot: 4 weeks, 20 seats, deployed into your contact-centre stack. We report AHT, CSAT, and script compliance weekly. You own the stack — and the data — before, during, and after.

20 seats · 4 weeks · weekly AHT / CSAT / script compliance report · on your infrastructure