On-premise, private cloud and sovereign voice AI

On-premise and private-cloud voice AI.

The full conversational AI stack running inside your data centre or your cloud account: STT, LLM, TTS, telephony, analytics and dashboards. No call audio or transcript leaves your network. Same product and APIs as EnableX cloud, so you can start there and move later.

Talk to our security team View the architecture

Cloud · Hybrid · Private cloud · On-premise · Air-gapped · Controls aligned with ISO 27001 and SOC 2 Type II

Deployment

Four ways to run it. One product.

EnableX voice AI can be deployed fully on-premise: speech recognition, the language model, speech synthesis, telephony adapters, conversation intelligence and dashboards all run inside your data centre, with no call data leaving your network. The same product also runs on EnableX cloud, in your cloud account, or in a hybrid split, and you can move between them without rebuilding.

Cloud (multi-tenant)

Start: Fastest, per-minute pricing

Data: In your chosen region

Best for: Pilots and non-regulated workloads

Hybrid

Split: Call media in your data centre, orchestration on EnableX cloud

Pricing: Annual plus per-minute

Best for: Telcos and banks with a carrier or PBX on-site

Private cloud

Where: Single-tenant in your AWS, Azure or GCP account

Control: Your keys, your audit logs

Best for: Banks, insurers, healthcare

Full stack

On-premise

Where: Your data centre

Pricing: Annual licence

Best for: Banks, government, defence; air-gapped option

Start anywhere, move anywhere, without rebuilding. Same product, same APIs, same dashboards.

The bundle

What is in the on-prem bundle.

Stack

Everything you need, nothing that calls home.

Streaming STT, an LLM runtime with open-source models packaged, streaming TTS, SIP and PSTN adapters, conversation intelligence, dashboards and an admin console. Bring your own key for a hosted model through your VPC if policy allows.

Streaming STT: open-source and licensed engines, self-hosted
LLM runtime: open-source models packaged, bring-your-own-key optional
Streaming TTS with bring-your-own voices
Conversation intelligence, dashboards and admin console

Sizing

Sized to your concurrent call load.

Reference architectures by concurrent calls, with GPU-backed inference for STT and the LLM and a CPU-only option for low volume. Bring your own hardware or use our reference specifications.

Reference architectures by concurrent call load
GPU-backed inference for sub-second turn-taking
CPU-only option for low-volume deployments
HA and DR configurations included

Security

Audit-ready out of the box.

Encryption at rest and in transit, role-based access with SSO, and an audit trail of every call, model invocation and data access. Audit reports support your regulator reviews. Controls aligned with ISO 27001 and SOC 2 Type II.

Encrypted at rest and in transit
Role-based access with SSO (SAML, OIDC, AD)
Audit trail of every call, model invocation and data access
Air-gapped option available

See full security posture

Operations

Updates on your schedule. Support that knows your stack.

Signed update bundles on a quarterly cadence that you choose when to apply, with out-of-band security patches when needed. 24x7 enterprise support, and a named technical account manager for large deployments.

Quarterly signed updates, applied when you decide
Out-of-band security patches
24x7 enterprise support with SLA-backed response
Named technical account manager for large deployments

Who needs it

Where deployment choice decides the deal.

Regulators in several markets require customer voice data to stay in-country or in-network. For these buyers, private cloud or on-premise is not a preference but a condition of the purchase.

Banking and insurance

Banks and insurers

Customer voice, collections calls and claims recordings kept in-country or in-network, as the regulator requires. Hybrid where the carrier or PBX is already on-site.

Banking voice AI

Healthcare

Hospitals and providers

Patient conversations handled under health-data rules and hospital data governance. On-premise or private cloud, hybrid where the EHR lives on-site.

Healthcare voice AI

Government

Government and public bodies

Citizen services on sovereign or air-gapped infrastructure, with no outbound connectivity and a full audit trail.

Government voice AI

Telco

Telcos

Voice AI offered to enterprise customers under the telco's own brand, carrier and compliance footprint, including customers who need on-premise.

Telco white-label

Deployment

What a deployment looks like.

Technical install is measured in weeks. Most of the elapsed time is your security review, network configuration and integration testing, so we front-load the documents your security and network teams need, and size the environment with them before anything is shipped.

Professional Services →

Supplied up front

Deployment runbook for your infrastructure team
Security questionnaire response for your security review
Reference architecture sized to your concurrent call load
Integration plan for telephony, CRM and contact centre
Acceptance tests run on your own numbers before go-live

Sovereign AI

Sovereign AI, in practice.

Sovereign AI means every layer of the AI stack, from model weights to telephony media and logs, runs on infrastructure you control. No data egress to a third-party cloud, no model calls to an external API, and no dependence on a vendor's logging for your audit trail.

Voice is the hardest layer to make sovereign, because STT, TTS and the LLM all have to run locally at conversational speed. EnableX ships that as one package, with audit logging your data protection officer can sign off on. Country-specific rules for India are covered on the Voice AI for India page.

Questions security teams ask about on-premise voice AI

Yes. Speech recognition, the language model runtime, speech synthesis, telephony adapters, conversation intelligence and dashboards all run inside your data centre. No call audio or transcript leaves your network. Cloud, hybrid, private-cloud and air-gapped options use the same product.

Open-source models such as Llama, Mistral and Qwen are packaged and run on your hardware. If your policy allows it, you can instead bring your own key for a hosted model from OpenAI, Anthropic or Google through your own VPC. EnableX provides the orchestration; the model choice is yours.

Technical install is measured in weeks. Most of the elapsed time is your security review, network configuration and integration testing. We supply a deployment runbook, a security questionnaire response and a reference architecture up front, and size the environment with your team.

Signed update bundles on a quarterly cadence. You choose when to apply them. Security patches are delivered out-of-band when needed. Nothing updates automatically, so you stay in control of what runs in your environment.

Yes. For government, defence and other sovereign workloads the stack runs with no outbound network connectivity. Updates are delivered on approved media, and every call, model invocation and data access is logged for your audit team.

An annual licence based on concurrent capacity, languages enabled and modules in use. No per-minute charges. Quoted after a security review and sizing.

Talk to our security team, not a salesperson.

Bring your security questionnaire, your compliance constraints and your network diagram. A security architect will walk you through what the deployment looks like in your environment.