Skip to content
The platform

Two products that share one set of foundations

Retrieval, connectors, tenancy, access control and deployment are built once and used by both modules. This page is the part that does not change when you switch channels.

Shared foundation

Two front doors, one engine room.

Whichever module you start with, the knowledge, connectors, tenancy and governance underneath are the same. Adding the second channel is configuration, not a second integration project.

Intelina Voice

AI voice agents + customer CRM

Intelina Desk

AI chat + ticketing tool

Shared platform layer
Knowledge retrieval·Live connectors·Tenant isolation·RBAC + audit·Encrypted secrets·Containerised runtime

One retrieval layer

Both modules index documents into the same search layer, partitioned by tenant. Upload once, ground answers on the phone and in chat.

Your systems, called live

Voice tools and chat data endpoints are the same idea in two channels: you register an API, the AI decides when to call it, and the response comes back fresh.

Multi-tenant to the row

Every query, cache key, search filter and API key is scoped to an organisation or project. There is no unfiltered fallback path in either module.

Governed by default

Role-based access, audit logging, encrypted secrets and per-key scoping ship with the platform rather than as an enterprise add-on.

Containers you can host

Every service is a container. Run it on our managed cloud or inside your own — the deployment topology is the same either way.

Adopt one, add the other

The modules are independent products with their own dashboards. Start with the channel that hurts most and turn the second one on later.

Side by side

Same platform, different jobs.

If you are trying to work out which module you need — or whether you need both — this is the honest comparison.

Comparison of Intelina Voice and Intelina Desk across twelve dimensions
DimensionIntelina VoicePhoneIntelina DeskWeb chat
ChannelInbound phone callsChat widget on your website
How customers reach itThey dial your numberOne script tag, no rebuild
What the AI isSpeech-to-speech agent with a persona, voice and toneText agent governed by a per-project AI policy
Identifying the customerCaller ID matched against your imported recordsYour own site tells the widget who is signed in
Where customer data livesImported into the platform as customer recordsFetched live from your APIs, cached briefly, never stored
KnowledgeDocuments indexed and retrieved per organisationDocuments indexed per project, tagged by category
Taking actionTools you define: webhook, lookup, escalate, end callRegistered data endpoints the model calls when relevant
Human handoffEscalation tool invoked mid-callConversation becomes a ticket with the full thread attached
Work managementCall logs, transcripts, detected intents, recordingsTickets, queues, SLA tracking, approvals
Team rolesOrganisation members with shared agent configurationSuper admin, project admin, supervisor, support agent
Real-time transportTelephony media stream bridged to the model socketLive updates pushed to agents and customers
Priced onMonthly call volumeFree tier, then plans by team and volume

Running both? Knowledge uploaded for one module can be reused by the other.

Runtime

What actually happens between the question and the answer.

No hand-waving: this is the sequence each module runs, in order, on every conversation.

What happens when the phone rings

  1. 01

    Call arrives

    The telephony provider posts the dialled number to the orchestrator, which resolves it to your organisation.

  2. 02

    Context assembled

    Agent persona, the caller's customer record and the relevant knowledge chunks are fetched in parallel and composed into one prompt.

  3. 03

    Agent connects

    The speech model opens with your persona, voice and the caller's real data already loaded.

  4. 04

    Conversation runs

    Audio is bridged both ways while transcripts are scanned for actionable intent.

  5. 05

    Actions execute

    A detected intent dispatches your tool, and the result is folded back into the agent's context.

  6. 06

    Call closes

    Remaining actions run, the transcript and recording are logged, and the caller receives an SMS confirmation.

Deployment

Run it our way, or entirely your own.

The reason the voice module runs its own speech model rather than renting one is exactly this: self-hosting has to remain a real option, not a slide.

Managed cloud

We run it. You get dashboards, a phone number or a script tag, and an onboarding session. Fastest path to a working agent.

  • No infrastructure work
  • Rolling updates
  • Shortest time to live

Your cloud

Every service is a container with an explicit configuration surface, so the same topology deploys into your own project or account.

  • Data stays in your perimeter
  • Your networking and IAM
  • Same services, same versions

Hybrid

Keep the parts that touch customer data inside your boundary and let us operate the rest — common where the voice module needs hardware you would rather not manage.

  • Split by sensitivity
  • Bring your own search index
  • Bring your own telephony account
Packaging
Containerised services, one image per component
Scaling
Horizontal, per service — the audio bridge and the API scale independently
Storage
Managed databases, with caching and queues handled for you
Search index
Tenant-filtered on every query
Secrets
Platform secret manager plus per-organisation encrypted values
Speech runtime
Dedicated hardware for the voice module; the chat module needs none

Put both channels on the same brain.

Bring one line or one chat widget. We will show you the same agent answering with your real data, calling your real systems, and handing off cleanly when it should.