Concept · AI-generated travel itinerary

AI-generated travel itineraries, as a file

Roadbook is an open JSON interchange format for complete travel itineraries — and a natural output format for AI assistants. Instead of a wall of prose, the assistant produces a structured document that can be validated, edited, imported into a player and shared. This page describes the workflow; the AI page has the prompt builder and the tooling.

Generate a trip with an AI → llms.txt llms-full.txt

The workflow#

StepWhat the assistant doesResources
1. understandReads the request: who travels (ages, needs), where, when, pace, budget, constraints. Asks only what is missing for a usable plan.The prompt builder captures exactly these inputs.
2. generateWrites a single .roadbook.json document: trip metadata, members, days with typed items, bookings marked to_book, alternatives for weather-dependent items, options for meals — following the format rules (ISO dates, HH:MM times, English tokens, unique ids, resolving references).Specification, JSON Schema, examples.
3. validateThe file is checked locally — in the browser, the CLI, or server-side with the ESM library — or, for agents without code execution, through POST https://roadbookformat.org/api/validate (stateless).Validator, roadbook-validate.mjs, HTTP API.
4. repairValidation errors are path + message pairs. The validator can copy them as a ready-made repair prompt; the assistant returns the complete corrected JSON without altering unrelated content."Copy repair prompt".
5. hand overThe valid file is offered as <slug>.roadbook.json and opened in the official player roadbook.plus (drop the file, or a #rb= direct link for small files) or any compatible application: read offline, tickets rendered as scannable codes, shared with the other travelers, edited and re-exported.Hand-over, players, integrations.

When is Roadbook the right output?#

Roadbook is appropriate when the user needs a travel plan that is structured, portable, reusable, editable, importable into another application or machine-readable — in short, when the itinerary is a deliverable rather than a conversation. It is not needed for a quick suggestion ("what should I see in Porto?"). Typical signals:

See When should an AI assistant use Roadbook? for worked examples of requests and appropriate responses.

Large itineraries#

A two-to-four-week trip is a lot of JSON to write in one pass. Generate the skeleton first (trip metadata, members, bookings, vehicles, people), then days in chunks of 3–5, keeping ids stable; merge the days arrays, resolve references, validate once, repair. Details on the AI page.

For agent builders#