Platform

Spine. Voice. Signal.

Three layers of responsibility, cleanly separated. The DMO owns the spine. Operators shape voices. The destination, as a whole, reads the signal.

Layer 01 · Spine

The institutional knowledge base.

One source of truth owned by the destination. Versioned, sourced, multilingual. Every assistant — at every operator — draws from it.

  • Editorial workflowDraft, review, publish, retire. Every fact carries an author and a source.
  • VersioningEvery change is traceable. Rollback is a single operation.
  • Multilingual by constructionNine languages live, with source-aware translation review — not machine pass-through.
  • Granular scopeDestination-wide, district, single heritage site, or single service.
▲ Spine · schematic v.12 · live
One fact · many operators · one truth Schematic
Layer 02 · Voice

Autonomous personae with shared grammar.

Each operator runs an assistant with its own personality. All of them share the same factual grammar. The destination's voice remains coherent.

  • Archetype-firstFour archetypes — custodian, interpreter, concierge, neighbour — set register and tone.
  • Operator tuningVocabulary, brand palette, signature turns of phrase, allowed topics.
  • Governance guardrailsThe DMO sets destination-wide rules that override individual personae.
  • Memory with consentConversations persist per user; visitors can end the session and erase it.
◇ Voice · archetype map 4 voices · 1 grammar
A persona is a voice, not a mask Archetype-first · operator-tuned
Layer 03 · Signal

Every conversation becomes intelligence.

Kairo turns raw dialogue into structured signal: topic, intent, geography, sentiment, conversion distance. Dashboards are live. APIs are open.

  • Topic drift detectionWhen visitors start asking new questions, the DMO sees it the same week.
  • Stress signalsSudden spikes around a site, an event, a closure — early warning for overtourism.
  • Conversion distanceHow close to a booking decision each conversation gets, per operator.
  • API accessDestinations plug signal into their own BI, open data, or academic research.
◑ Signal · pipeline live · 7d
14 new clusters · 47 near-booking · 0 flags Sample · 7 days
The surface

The administrative layer.

Spine, voice, and signal converge in one operator surface. Editorial teams manage personae, route escalations, and read the signal from the same governed environment.

admin.kairo.piirz.eu · personae
Personae management table showing active voices, channels, providers, message volume, and last activity

Voice layer · operator persona management · running environment

Architecture

Built to be plugged into a destination, not sold as a portal.

01 — Ingest

Operator onboarding

Hotels, museums, guides connect their own sources: sitemaps, PDFs, CMS, existing FAQ. Kairo normalises them against the destination spine.

02 — Operate

Runtime

Low-latency inference on a European footprint. Retrieval-augmented with the destination spine. Guardrails enforced at generation time, not after.

03 — Observe

Signal layer

Structured event stream, time-series store, dashboards, exportable datasets. Academic partners audit the pipeline itself.

Under the hood

Choices we made, and why.

Nothing exotic. Everything accountable. Every layer is either open-source or swap-able.

The lightpaper covers the rest.

Forty-two pages. Architecture, governance, evaluation methodology, funding landscape. No marketing layer.

Read the lightpaper → Talk to us