Vaifredo

The wager

It's a Friday night in São Paulo. Four friends are flying down from Salvador in the morning to spend the weekend in Trancoso with you. You've already booked the house in Bahia. They've already booked their flights. Someone forwarded a Google Doc with a list of restaurants. Someone else made a WhatsApp group. The host is in three different threads about the same trip.

You open Vaifredo. It knows the four of you are arriving Saturday. It knows the house. It knows the airport pickup time. It knows the group's prior trip preferences — Mocotó-loud over D.O.M.-fancy. You ask: "What should we do Saturday afternoon?" It answers in two options, both near the house, both group-sized, both timed to land before the dinner reservation it remembers you mentioned in last week's chat.

Vaifredo is a bet that travel apps fail because they make you re-explain yourself every time. Vaifredo never asks. It already knows.

Every other AI travel tool starts the conversation from scratch. Where are you staying? Who's coming? What did you book? Vaifredo treats your trip as a context — bookings, group, location, history, decisions already made — and answers accordingly. The chat is downstream of the data.

Google Trips died. Wanderlog is a CRUD app with a map. TripIt parses emails but doesn't reason. Gemini's travel features know nothing about who you are or where you're staying. Vaifredo is what travel planning looks like when the AI starts already knowing.

How it's built

The product surface is a PWA — installable on a phone home screen, offline-capable for itinerary access, no app store gatekeeping. Map-first, list-second. Mapbox handles the rendering; Claude handles the reasoning.

Three pieces are doing the load-bearing work.

Email-import-as-onboarding. New users don't fill out forms. They forward a confirmation email — a flight, a hotel, an Airbnb — to a Vaifredo inbox. The parsing pipeline reads the structured data, infers the trip dates, the destination, the party size, and creates a draft trip workspace. Subsequent forwarded emails get attached to the same trip automatically. The user's first interaction is receiving a populated trip, not building one. Onboarding takes one email.

Local context cards. When you're inside a trip, the chat has access to a layered context: your bookings, your group's preferences, the time of day, your current geolocation if you're on-trip, and the map's current viewport. When you ask "what's good near here?", Vaifredo doesn't just answer with restaurants — it answers with restaurants that are open, that match your group size, that fit the window between your current location and your next reservation, that you haven't already been to. Each suggestion is a context card with the proof — the timing, the distance, the why — visible underneath. Nothing in the answer is unsourced.

Proactive nudges. Some moments deserve interrupting the user. Your flight is delayed and the airport transfer needs rescheduling. The restaurant you booked has a 30-minute wait and your group is still 20 minutes away. The weather just turned and the beach day needs to become a museum day. Vaifredo watches the trip context and pushes timing-aware notifications when the trip itself has changed. The bar for a nudge is high — most travel apps spam; Vaifredo only speaks when the trip needs it to.

The shared workspace layer is the social piece. Everyone in the group sees the same trip, the same map, the same decisions. Changes show whose change. Disagreements happen in the chat, not in three parallel WhatsApp threads.

What's working

Vaifredo is in private beta. Deployed, usable, in active use across a small set of real trips — group weekends in Brazil, a multi-country Europe trip, and a Bahia long-weekend that hit every one of the proactive-nudge use cases.

What's working in practice:

• The email-import onboarding eliminates the cold-start problem. Users arrive with a trip already populated; the chat starts useful.
• The local context cards consistently beat "generic restaurant recommendations" because the answers honor the constraints the user didn't have to state.
• Proactive nudges have a high signal-to-noise ratio when the threshold is set conservatively. The trips where nudges fired, the nudges were the right calls.

What's not yet proven:

• The shared workspace layer hasn't been stress-tested with truly large groups (8+). The conflict resolution UX may need work.
• The email parsing pipeline is reliable on major airlines and hotel chains but struggles on smaller Airbnb properties and Brazilian booking platforms that don't follow standard confirmation formats.
• The cost economics of a heavily-context-aware chat are unproven at scale. Every Vaifredo query loads more context than a generic chat — token cost per session is higher than the comparable Pico session.

What's next

Public beta is the next milestone. The gating decision is whether to lead with English-first international travel or Portuguese-first Brazilian travel. The architecture supports both, but the marketing posture, the seed corpus of partner properties, and the early-user acquisition channels look different in each direction.

The bigger architectural question is whether Vaifredo should ingest more deeply — direct integrations with airlines, hotels, calendar apps — or stay on the email-forward + manual-entry model. The deeper integration is more powerful and more lock-in-y. The lighter model is more sovereign and more universal. There's a real philosophical call to make.

Vaifredo is the most ambitious thing under the Frenti roof — the only product that requires being right about the user's whole trip, not just the next question. Getting context to flow correctly across that surface is the hardest engineering problem in the portfolio. It's also the one where being right matters most.