Large specs and performance

How Apifolio handles very large OpenAPI files such as GitHub's and Stripe's REST APIs, with measured numbers and limits.

Apifolio is tested on some of the largest public specs: GitHub's REST API (a 10 MB YAML file with 1,232 endpoints) and Stripe's (an 8.5 MB JSON file with 612 endpoints). Performance budgets run with the automated tests, so a change that makes these specs slower fails the build.

How it stays fast

  • Files larger than 400 KB are parsed in a background thread (a web worker), not on the page's main thread.
  • Endpoints are rendered as you scroll, and schemas are resolved when you open them.
  • Search indexes are built while the browser is idle, so the first search is quick.
  • Macros that show the same file on one page share one download (in browsers with BroadcastChannel and Web Locks, which current browsers have).
  • PDF and Word exports of a page with many macros of one file reuse one download and one parse while the export function stays warm.

Measured numbers

Measured on 7 October 2026 on the code of Apifolio 2.2, on a desktop PC with an AMD Ryzen 9 9950X3D and Windows 11. Your numbers depend on your computer and network.

Step GitHub (10 MB, 1,232 endpoints) Stripe (8.5 MB, 612 endpoints)
Parse the file and build the API model (Node.js 24) about 190 ms about 16 ms
Export the largest section for PDF or Word about 13 ms (203 endpoints) about 23 ms (47 endpoints)
Build the Confluence search digest about 8 ms about 2 ms
Main-thread blocking while the macro opens (headless Chrome, CPU slowed down 4 times, median of 5 runs) 0.5 s 0.4 s

The Node.js figures come from the performance tests, run on their own three times. Main-thread blocking is the Total Blocking Time: the sum of every stretch longer than 50 ms during which the page could not respond. A synthetic 24.9 MB YAML file, close to the 25 MB limit, measured 2.1 s, with a longest single pause of 1.2 s, in the same conditions.

In the export tests, a page with 25 macros of the GitHub spec (one per section) takes about 0.4 s of processing in total, with one download and one parse of the file in a warm Forge function.

Limits

  • Spec files up to 25 MB.
  • Up to about 4 MB for anonymous visitors (see Public spaces).
  • PDF and Word exports: full details for up to 250 endpoints per macro (see Export).

Tips for very large APIs

  • Use several macros, each showing a few sections (tags) of the same file. They share one download, and each one indexes its own endpoints for Confluence search.
  • Choose a fixed height so the page stays easy to scroll.
  • Use search: it covers every endpoint and data model the macro shows, even the ones not on screen.

View as Markdown