Pillar 01 — Safari AI

Read the enquiry.
Draft the proposal.

The work between “WhatsApp pinged” and “proposal sent” is the operator’s most expensive hour of the day, and most of it is re-typing what the enquiry already said. Safari AI reads the enquiry, drafts the skeleton, costs it against your rate cards and your accepted-quote history, and writes the day-by-day. You read it. You edit it. You send.

Three passes. One screen. Always streamed.

01

Brief

Brief extractor

Haiku reads the enquiry — WhatsApp message, voice note, forwarded email, advisor PDF — and pulls out the structured brief. Party composition, dates, budget band, lodge tier, must-see destinations, dietary needs, mobility constraints. Every field shows the confidence score; anything the model isn't sure about gets a chip you tap to edit before generation runs.

02

Skeleton

Skeleton planner

Haiku again — fast, cheap — sketches the trip. Country sequence, night counts, lodge tier per stop, key activities, internal-flight legs. You see it before the costed pass begins. Reject the skeleton if it's wrong; tell the model why; regenerate.

03

Costed full pass

Costed full pass

Sonnet completes the itinerary with line-item pricing against your rate cards. Park entry, conservancy levies, gorilla permits, bush-flight tariffs, single supplements, FX from supplier currency to your sell currency. Verify-before-send chips appear on line items the model couldn't ground in your data.

Streamed, not batched

The brief and the skeleton land in seconds. You read the model's work while it's working, not after a spinner. You can intervene at any pass.

Honest about uncertainty

The model flags what it isn't sure about. FX rates older than your threshold, line items not grounded in your rate cards, ambiguous brief fields — all surfaced as chips on the proposal before you send.

PII never leaves your org

Traveller names, emails, and phone numbers are scrubbed before any AI provider sees the request. The operator side of the proposal has the names; the prompts don't.

Trained on the quotes your operators already accept.

Every accepted itinerary feeds the retrieval layer. Safari AI ranks lodge picks by similarity to your past accepted bookings, weighted by recency. Your accepted-quote history stays inside your organisation; nobody else’s Safari AI sees it.

See how The Bush feeds it