June 10, 2026 · 7 min read · Data

Corrected 14 November 2026: the example responses below are what the API returns today; the earlier version showed made-up fields (eligibility, max_stay_days, as_of) and a value, not_admitted, the data does not use.

The visa matrix is a graph, not a country list.

"Does a Brazilian need a visa for Japan?" is not a fact about Brazil and not a fact about Japan. It's a fact about the edge that connects them. Visa rules live on a 200×200 graph of passport-by-destination pairs, and treating them as anything else produces hallucinations that have real consequences.

Why generic LLMs get this wrong

A frontier model has seen "Schengen allows visa-free travel for 90 days" a thousand times in training data. It has not seen "South Korean passport holders entering Croatia after the 2023 Schengen accession" with the same frequency. So it confidently pattern-matches:

The cost of being wrong is not abstract: a user denied boarding because your AI itinerary said "no visa needed" doesn't get a refund. The matrix has to be ground truth.

The shape of the matrix

/v1/visa returns one row per (passport, destination) pair:

GET /v1/visa?passport=DE&destination=TH

[
  {
    "passport_country": "Germany",
    "destination_country": "Thailand",
    "requirement": "visa_free",
    "duration_days": 60,
    "notes": null,
    "license": "MIT",
    "source": "Passport Index",
    "source_url": "https://github.com/ilyankou/passport-index-dataset",
    ...
  }
]

The five requirement values in the data are:

Schengen, Five Eyes, and the common-passport patterns

A few clusters dominate the matrix:

These blocs make ~40% of the matrix predictable. The other 60% is where the work is.

Use cases that need the edge, not the country

Once visa is data, you can do things you couldn't do before:

  1. Filter recommendations. "Suggest weekend trips from Dubai for an Egyptian passport" — exclude any destination where eligibility is not visa_free or VOA with short turnaround.
  2. Gate UX flows. Don't show "book your trip to Cuba" CTAs to a US user without an accompanying disclosure about the OFAC categories.
  3. Itinerary feasibility. A 14-day Europe trip for a Filipino passport with a single Schengen visa works; the same 14 days hopping in and out of UK + Schengen requires two separate visas.
  4. Cost estimation. Visa fees are part of the trip budget. Sum them at planning time, not at checkout.
  5. Risk warnings. If eligibility is not_admitted, the entire trip is non-viable; surface that before the user spends an hour planning hotels.

Freshness and the as_of field

Visa policy changes faster than most travel data. Thailand extended visa-free stays in 2024, Indonesia rolled out its new "Golden Visa" the same year, the UK switched to ETA in 2025, the US suspended the visa interview waiver pilot. Each visa row carries its source and licence, but no date of its own, so check the source before your application relies on a rule.

For high-stakes flows (booking, compliance), pair the matrix response with a "verify with the embassy" disclosure. The matrix is the best fast answer; it isn't a legal opinion.

Sign up — free See India coverage →