Homebrew cask installs: a browser count that never reads a user-agent

The first figure in this spoke for brave. Homebrew is the dominant third-party package manager on macOS; it counts every brew install of a cask and publishes the aggregate as open JSON. Because the count happens at the package manager, the browser is identified by the name of the thing being installed, not by a string the browser sends to a website. That is precisely the instrument growth edge 9 asked for, and it is the reason Brave appears here at all.

Snapshot read 2026-08-21, covering 2025-08-21 to 2026-08-21: 25,137,516 cask install events across 22,411 casks.

The browsers

Brave's beta and nightly channels add 1,141 and 1,072; brave-origin, the build without Rewards, adds 478.

What it says that no user-agent panel can

Brave outranks Edge here, 1.8 to 1. On every panel that reads the user-agent, Edge is a mid-single-digit share and Brave is not a row at all. dap-browser-share breaks out 31 browsers across forty million federal-site users, and Brave is absent from the list. The privacy design that makes Brave invisible to those instruments does nothing to a package manager, so the ordering flips.

The two instruments also disagree violently about the small browsers, which is a warning about reading either as a population. DAP has opera at 144,948 users and vivaldi at 125 — Opera ahead by a factor of about 1,160. Homebrew has Vivaldi ahead of Opera, 18,363 to 9,854. Both numbers are honestly collected and they describe different populations, so neither is "the" share. What can be said is narrower: inside the macOS package-manager channel, Brave is the largest non-incumbent browser by a wide margin.

What it is not

• Install events, not people. A reinstall, a second machine, a rebuilt CI image and an upgrade path all add. The payload carries no user identifier and no IP field, by design, so nothing can be deduplicated to a person. This is a floor on activity, not a user count.
• macOS only, and developers. The most-installed cask in the window is claude-code at 1,107,061 (4.40%), with codex second and docker-desktop fourth. That is the population being measured, and it is not the public.
• One distribution channel. Most Brave users on macOS take the .dmg from brave.com; those installs are invisible here. The ratios above are shares of Homebrew, and cannot be scaled up.
• Opt-out, and the opt-out is correlated with the subject. brew analytics off and HOMEBREW_NO_ANALYTICS=1 both disable collection. The users most likely to set them are the privacy-minded ones, the same population that chooses Brave. The bias runs against the number reported here, so 93,290 should be read as a lower bound within its own channel.

Tier

T1. Homebrew is the operator of the channel and publishes both the aggregate and its collection policy; the claim tiered here is "how many Homebrew cask installs happened", which is a fact about Homebrew's own system, reported by Homebrew, with the retention (365 days in InfluxDB) and the opt-out documented. It is a first-party fact from a party that sells no browser.

Connections

• the browser it was hunted for: brave
• the instrument it corrects: dap-browser-share · statcounter-browser-share
• other browsers it counts: vivaldi · opera · firefox

A second package-counting channel (added 2026-08-22). brave-play-store-listing does on Android what this panel does on macOS — counts the package, so the Chrome user-agent is irrelevant — and it is bucketed to 100M+, so it confirms the approach without sharpening the figure.

Related

brave · dap-browser-share · statcounter-browser-share · vivaldi · opera · synthesis