Simple Host for Travel

Plan in progress · October 2026

Simple Host for Travel

Open-source software for small travel agencies. Paste a WhatsApp enquiry and get a branded quote page on your own domain in minutes. When the client says yes, they get a voucher and a live trip page for the trip itself.

Open source, Apache-2.0 Runs on Simple Host or your own server

This page describes the plan. The product is being generalised from a system a travel agency in Delhi has run live since September 2026, and it has not been released yet.

What every client gets

Four links, all on your own domain

Quote

Days, hotels, photos, and the price, with your logo. Every change keeps the same link.

<client>.quotes.<your-domain>

Voucher

Confirmation numbers, contacts, transfers, and flights to show along the way. No prices.

<client>.voucher.<your-domain>

Trip page

What's next today, a map, and the tickets, behind a 6-digit code.

<code>.trips.<your-domain>

Destination guide

When to go, visa basics, the main places, and sample itineraries, to share before anyone books.

<place>.guide.<your-domain>

The agency owner's day

Your day, from enquiry to trip

You run a small travel company, maybe with a team member or two, and your clients find you on WhatsApp. Here is one booking, start to finish.

You stay on WhatsApp the whole time. The quote, the voucher and the trip page are made for you, and your client keeps the same links from the first message to the last day. The dashed arrow is a change: as many as the client wants, always on the same link.

And at any time

See what your team did

Every quote, change, and ticket, with who did it and when. Each person signs in as themselves, and nobody can delete the log.

Set your rules once

Currency, markup, the tax lines for your country, per-person pricing. Every quote follows them, whoever makes it.

Your logo on everything

Upload it once. Every page carries it, and every PDF has it as a faint watermark.

Your client's trip

The same booking, from your client's side

Your client never downloads an app or makes an account. Imagine you are them.

Every link arrives on WhatsApp and opens on any phone. Once opened, the trip page keeps working when the signal drops, and the agent is always one tap away.

Why it works

Why agencies like it

Quotes that look like a travel brand, in minutes

Paste an enquiry and get a page with your logo, photos, a day-by-day plan and the price, ready to send.

One link that stays up to date

Every change lands on the same link, so nobody has to ask which version is the latest.

Everything a traveller needs on their phone

What's next today, the map, the tickets and a button to reach you, even when the signal drops.

Your rules, applied every time

Set your currency, markup, and taxes once, and every quote follows them, whoever on the team makes it.

Nothing gets lost

Every version of every page and every action by your team is kept. Deleting only archives.

Works for a one-person agency

No developer and no designer. Team members can make and send quotes without touching your templates or rules, and you choose what else they may do.

Your data on your server

The console and your client data run on your own server. The pages live on Simple Host under your own domain, hosted or on the same server, and it is the same product either way.

About 2¢ of AI a quote

Paid straight to the AI provider you choose, plus one small server. Or connect ChatGPT, Claude, or your own agent and use the plan you already pay for.

Any country

Built first for India, with rupees, GST, and TCS. Pick your country at setup and the currency, taxes, and language follow, each changeable in a setting.

Open source

Apache-2.0. The code is public, the installer is one command, and an upgrade is the same command again, with a backup first and a rollback if anything fails.

Cost

About 2¢ a quote

You pay your AI provider directly. The default model is deepseek/deepseek-v4.1-flash-1017 on OpenRouter, at OpenRouter's prices on 20 October 2026. Most of what the assistant reads at each step is the same page and the same rules, which the provider caches.

Input $0.15 per 1M tokens Cached input $0.003 per 1M tokens Output $0.60 per 1M tokens
One new quote, about 10 assistant stepsTokensCost
Input read from the cache450K$0.00135
New input50K$0.0075
Output20K$0.012
Per quote≈ 2¢
100 new quotes a month≈ $2.10

A change to an existing quote is a smaller run than a new one.

Gemini free tier

No cost within Google's free limits, good for trying it. Google may use free-tier prompts, so not for real client data.

Ollama Cloud

For agencies that want a provider that does not log prompts.

The server

One small VM. The console is a single program that runs in 512 MB of memory when the pages are on simple-host.app, and in 1 GB with your own Simple Host.

For the technical personUnder the hood What runs where, hosted or self-hosted, installing it, how the AI's work is checked, and the full feature list.

In these diagrams solid arrows move forward and dashed arrows go back a step. Shaded boxes are the parts this project builds; dashed boxes are services the agency picks.

What runs where

The agency runs one small server. The pages live on Simple Host, under the agency's own domain. The AI provider is whichever one the agency chose.

The console holds the rules, the log, the team accounts, and the trip data; Simple Host holds every page and every version. The model only sees prompts and tool results. Trip passcodes live on Simple Host, and team members copy them from the console, so they never reach the model.

Two ways to run it

The console is the same in both. What changes is where Simple Host runs, which decides what the agency looks after. Moving from one to the other keeps every page, version, and trip.

Option A · pages on simple-host.app

The agency runs only the console. Simple Host keeps the pages, versions, passcodes and certificates. DNS points the four address families at simple-host.app.

Option B · everything on your server

The console and Simple Host's small-box edition share one VM. The only thing that leaves it is the call to the AI provider, and with Ollama Cloud that call is not logged.

Installing it

The installer does the server part and proves the chosen model can make a quote. The rest happens in the console, as the admin. An upgrade is the same command again: backup, update, health check, and a rollback if anything fails.

If the test quote fails, the installer says why and asks for a different model or key before going further. A model that cannot pass the checks on a sample enquiry will not pass them on a real one.

How one AI request runs

A team member types a change. The assistant works in a loop until the job is done, and the console checks the model's work itself before a draft is saved and again at Publish.

Why the checks are on the server

Cheaper models made real mistakes at the Delhi agency. Instructions in a prompt are not enough, so the console refuses a bad save and sends the reason back to the model, which fixes it before anyone sees the draft.

Pricing maths

Per-person prices, the markup, and the tax lines in order have to add up to the total on the page. The calculator tool is there so the model does not do this in its head.

No placeholders

Unknowns stay [in brackets] while drafting, which is allowed. Publish refuses a page with brackets left and shows the list.

Required fields

A quote without the parts every quote needs, such as dates or the number of travellers, is refused.

If the provider fails

A slow or failing provider is retried, then swapped for a fallback model. A draft is saved whole or not at all, an interrupted task can be resumed, and a step limit stops a runaway loop and says what is left.

Everything it does

Everything without a Planned mark runs at the Delhi agency today.

Team members

Anyone the admin has added. Makes, revises, and sends.

  • New quote from an enquiry. Paste the text, or attach the WhatsApp message, PDF, Word, Excel, or photos. The assistant picks the closest template, fills it, and names it after the client.
  • Change it by chat. Type the change in plain words in the quote's own conversation. The draft updates and the client link stays the same.
  • Tap to edit photos. Swap, add, or move a photo from the phone's camera, its gallery, or the shared photo library. iPhone and WebP photos are converted and resized for the web.
  • Draft until Publish. Clients only see published versions. Every version is kept with who made it and why, and any one can be restored.
  • Status per quote. Draft, sent, revised, confirmed, travelling, done, or lost, one tap, dated.
  • Trip page and tickets. The assistant builds the trip page from the quote. Team members upload tickets per step, copy a ready WhatsApp message with link and passcode, change the passcode, or sign every phone out.
  • Voucher. Hotel confirmations, contacts, transfers, and flights at the client's voucher link, with no prices.
  • PDF of every page, with the agency logo as a faint watermark.

Admin

Usually the owner. Everything team members can do, plus the rules and the team.

  • One admin and team members. The admin chooses what team members may do: change templates, change house rules, publish, archive, read the activity log, manage trips. By default templates, house rules, the logo, archiving, and the activity log are admin-only, as at the Delhi agency. Planned
  • A username and password for each person. The activity log shows real names, and the admin can change one person's access without touching anyone else's. Planned
  • House rules the AI always follows. Currency, forex markup, the tax lines, per-person pricing, wording. Every version of the rules is kept.
  • Activity log nobody can delete. Who did what and when, in the console and through the assistant. Filter by person, page, or date, or ask the assistant.
  • Logo once. Upload it once; every page header and every PDF watermark uses it.
  • Archive, never delete. Removing a quote takes it, its voucher, and its trip page offline. Archived brings all three back as they were.
  • Templates and destination guides, made and changed by chatting with the assistant. Templates for popular trips and a sample guide come installed, so the first quote and guide take minutes. Planned
  • Download everything. One zip of every page with all its versions, all trips and tickets, readable without any software. A copy is also made every night, it can go to storage off the server, and one command restores an agency onto a fresh server. Planned for the off-server part.
  • Run it from ChatGPT, Claude, or your own agent. An MCP server gives the AI you already use the same tools as the console, signed in as a named person with that person's role, and a published skill tells it how to write quotes, vouchers, trip pages, and guides. This also lets an owner use the AI plan they already pay for. Planned

Traveller

The client, before and during the trip. Never signs in.

  • Links that open on any phone. Quote, voucher, and guide links need no app and no sign-in.
  • Trip page behind a passcode. Six digits, typed once per phone, kept until the agency changes it.
  • The next thing to do comes first, then a map with numbered stops and every step with one button: directions, ticket or call.
  • Tickets in the same place, behind the same passcode, plus WhatsApp and call buttons to the agency.
  • Works offline. The trip page and any ticket opened once stay on the phone.
  • Goes offline when the trip ends. A closed trip shows "This site is offline". The agency can open it again with the same code.

AI assistant

Writes and changes the pages. Never sees a passcode or a key.

  • Built into the console. A small assistant loop inside the same program, with no separate agent service to install or update. It plans, calls a tool, checks the result against the page and the house rules, fixes, and retries until the job is done, and asks one clear question only when it is genuinely stuck. Planned
  • Survives a bad day at the provider. A slow or failing provider is retried and then swapped for a fallback model, drafts are saved whole or not at all, an interrupted task resumes, and a step limit stops a runaway loop. Planned
  • Bring your own key. Any OpenAI-compatible provider. The default is deepseek/deepseek-v4.1-flash-1017 through OpenRouter, with caching always on. Planned
  • House rules on every conversation, sent again at the start of each one.
  • Checks on every save. The server checks the pricing maths and requires the fields every quote needs, and a refused save goes back to the model with the reason. [Placeholders] are allowed in drafts and refused at Publish, with the list shown.
  • A calculator tool, so totals, markups, and taxes are computed rather than guessed.
  • Nothing invented. Unknown prices, times, and phone numbers stay in brackets until someone fills them.
  • One conversation per page, shared by the whole team, plus one general conversation. One request at a time per page; different pages run side by side.

Agency setup

Done once, by the owner or whoever helps them.

  • One command on a small VM. The installer asks, in plain words, for the agency name, the web address, the country, where the pages live, the AI provider, and the key, then checks everything. Planned
  • Upgrade with the same command. Backup first, update, health check, automatic rollback if anything fails. The console says when a new version is out, and nightly automatic updates are optional and off by default. Planned
  • Your country. Currency, number and date formats, language, and phone code come from standard country data for every country, with hand-checked tax defaults for the fifty countries most of Simple Host's visitors come from. A tax line can apply only to overseas packages, be worked out on earlier lines, or be included in the price and not shown. Every value can be changed any time. Planned
  • Your own domain. Four address families, *.quotes, *.voucher, *.guide, and *.trips, each one wildcard DNS record plus a TXT record that proves the domain is yours.
  • Pick an AI provider. OpenRouter by default, Gemini's free tier to try it out (Google may use free-tier data), Ollama Cloud for no logging, or any OpenAI-compatible endpoint. Planned
  • A test quote before you start. The installer runs a sample enquiry through the chosen model and the same checks a real quote gets. Planned
  • Settings instead of code. Branding and colours, page style, buttons, and page types are settings in the console. Planned
  • Report a problem. One button files a public GitHub issue with the version, the screen, and the steps, never client data, and shows exactly what will be sent first. Planned
  • Hosted or self-hosted. Pages on simple-host.app under your domain, or on your own Simple Host on the same server. Same product either way, and moving between the two keeps every page, version, and trip. Planned for the self-hosted side.

Status

Where it stands

  • Running for one agency. Quotes, revisions, vouchers, trip pages, guides, roles, house rules, and the activity log, live in Delhi since September 2026.
  • Simple Host pieces it needs. Address families and site passcodes shipped on 29 September 2026.
  • Licence and shape decided. Apache-2.0, one admin and team members with their own accounts, four page types, a built-in assistant loop with no bundled agent, any country with region defaults, and two ways to run it. Decided 21 October 2026.
  • Public repository, with the agency's own names, templates, and settings taken out, and every agency value a setting.
  • Built-in assistant loop with retries, a fallback model, safe resume, and a step limit, chosen by running the same test jobs through the candidates.
  • One-command installer and upgrader, with the plain questions, the provider choice, and the test quote.
  • Settings screens for the country, branding and colours, page style, buttons, page types, and what each team member may do.
  • Team accounts, a username and password for each person, with real names in the activity log.
  • MCP server and the published skill for ChatGPT, Claude, and any MCP client.
  • Docs: install, first quote, house rules, running it yourself.
  • First public post, with the Delhi agency as the case study.