Radar Vectors

About Radar Vectors

Radar Vectors is an independent search engine for public FAA documents. It is not run by, endorsed by, or affiliated with the Federal Aviation Administration.

What is in the library

73,765 documents and 1,808,263 searchable passages, copied from the official sources below and re-checked on a schedule. Each document page links back to the official copy.

SourceDocumentsLast checkedNot searchable
Federal Register (federalregister.gov)34,622Nov 19, 2026—
FAA Orders & Notices (faa.gov)20,056Nov 19, 202655 scanned or unreadable files
FAA Dynamic Regulatory System (drs.faa.gov)9,500Nov 19, 2026—
Electronic Code of Federal Regulations (ecfr.gov)5,715Nov 18, 2026—
FAA Advisory Circulars (faa.gov)1,923Nov 19, 20264 scanned or unreadable files
FAA Chief Counsel legal interpretations (drs.faa.gov)1,068Nov 18, 20261 scanned or unreadable files
Safety Alerts and Information for Operators (faa.gov)594Nov 19, 20262 scanned or unreadable files
FAA Handbooks and Manuals (faa.gov)165Nov 19, 2026—
Aeronautical Information Manual and Pilot/Controller Glossary (faa.gov)78Nov 19, 2026—
Airman Certification Standards and Practical Test Standards (faa.gov)44Nov 19, 20262 scanned or unreadable files

How search works

  1. Understanding the query. Document numbers (91.213, AC 61-65, 8900.1) are recognised and opened directly. Misspellings are corrected against the words that actually occur in FAA documents, and pilot shorthand is expanded to the FAA’s wording.
  2. Finding passages. Every document is split into passages. A full-text engine ranks them, giving weight to titles, headings, exact phrases and words that appear close together, and preferring current, binding material over cancelled or advisory material when relevance is otherwise equal.
  3. Finding passages by meaning. Each passage of the regulations, circulars, orders, handbooks, AIM, test standards, legal interpretations and safety notices also has an embedding: a list of numbers that places passages with similar meaning near each other. Passages near the question are added to the candidates even when they share none of its words.
  4. Reading the best candidates. Jev, a model from TypeSafe AI that returns probabilities rather than text, looks through the library’s tables of contents for documents that fit the question. It then reads the candidates and judges whether each passage answers the question. For the leading documents it reads every excerpt on its own, and checks that the likeliest answers are about the same kind of pilot, aircraft and operation as the question. Only then is one shown as a quote, marked “may be relevant” when Jev is less sure. If you set Asking as (for example “Drone, recreational”), Jev reads every question in that light; the setting stays in your browser. Nothing shown here is generated text: every passage shown is a quotation from an FAA document, with its page number. It is not a ruling on your situation.
  5. Showing what explains a rule. Under a regulation in the results you see the circulars and handbooks that cite it and how many legal interpretations discuss it, taken from the citation links between documents.

Limits you should know about

Changes and alerts

Each update compares every document with the copy already here and records what changed: new documents, status changes such as a cancelled circular or a superseded directive, and text changes, with a before-and-after view for regulation sections and other short documents. Regulations also carry every amendment the eCFR has recorded since 2017, linked to the official comparison. What’s new lists them, every document page has a History box, and Atom feeds (for the whole list, a filter, a single document or your watch list) let a feed reader or feed-to-email service tell you when something changes. The watch list stays in your browser: there are no accounts, and this site does not record who follows what.

What this site records

To find out what people cannot find, the site keeps an anonymous log: the words of each search, its filters and “Asking as” setting, how many results came back, whether a quote was shown, which result was opened, and the answers to “Was this passage helpful?”. It does not record your IP address, browser, cookies or the page you came from, and times are rounded to the hour. Each search gets its own random id, so separate searches cannot be linked to each other or to a person. The log is deleted after 90 days. The privacy page has the details.

Feedback

To tell us what you could not find or what is wrong, use the feedback form. It goes to the people who run this site. It is not a way to ask for a ruling on your own situation.

For developers

A JSON API is available at /api/search?q=… and /api/suggest?q=…. Search leaves out cancelled and historical documents. Add old=1 to include them.