← Design · c130

research/claims.md

Claims checked

Use this file to answer, with sources, the things people have heard about GPUI before they come up in Tyler's session: the slogans, the big-name adopters, the memory numbers, and the state of the framework. Each claim gets what was said, what the sources say, and a verdict: holds, partly, does not hold or unverified. Every source was read on 2026-11-09; secondary blogs are named as such. The last section lists claims in our own documents that this reading retires.

The relayed claims

ClaimWhat the sources sayVerdict
"GPUI is the Electron killer"The phrase is a blog title: Dishant Sharma, dev.to, 2026-10-06, "GPUI by Zed: The Electron Killer That Never Left Home". Its figures (120 FPS; about 2 ms input latency against "12ms+" for Electron; 150 to 300 MB against 650 MB to 1 GB "for an Electron editor") are introduced as "the numbers people keep repeating" and carry no citation; the post links to no external source. Its own thesis is the opposite of the slogan: GPUI "belongs to an editor". The same slogan was applied to Tauri years earlier (Better Programming; Syntax episode 821). — https://dev.to/dishant0406/gpui-by-zed-the-electron-killer-that-never-left-home-209o, https://syntax.fm/show/821/is-tauri-the-electron-killerA headline, not a finding. Unverified as a claim about performance; no measurement stands behind it
"Discord rebuilt its app on GPUI"Announced, not shipped. On 2026-11-04 Discord's co-founder and CTO Stanislav Vishnevskiy posted on X that the desktop client moves off Electron and React to a Rust core with "a customized version of GPUI", not Tauri; iOS and Android keep native interfaces and share the Rust logic (React Native dropped); YouTube embeds and Activities stay in the system webview; an early Windows build with two accounts open showed about 290 MB of memory; "user testing soon", no date. The primary post (x.com/svishnevskiy/status/2118081920194203799) was not fetched; it is reported by heise (iX, 2026-11-06) and a dev.to analysis (2026-11-08) that calls the screenshot "a real data point", "not a controlled benchmark". The codename "Discord Phoenix" appears only in secondary blogs. — https://www.heise.de/en/news/Discord-is-rebuilding-in-Rust-leaving-React-and-Electron-behind-11516113.html, https://dev.to/jamilxt/electron-vs-gpui-what-discords-290-mb-rust-rewrite-can-and-cannot-prove-34b6Holds as an announcement, five days old. Nothing shipped; one screenshot, no benchmark, no Electron baseline published
"Spotify rebuilt its client on GPUI"No source. Spotify's own client is unchanged. What exists are community clients written on GPUI: Cadence (macOS, by Miho), rspotify (OscarArranz), Spotirust, Riddim (a Linux port of Cadence), Natify, rust-player; Spotifast is egui, not GPUI. — https://github.com/OscarArranz/rspotify, https://github.com/Letdown2491/riddim, https://github.com/crmne/spotifastDoes not hold. Most likely a conflation with the hobby players
"A GPUI app idles at 150 to 300 MB; Zed uses a fraction of VS Code"Zed idle, from review sites that state no method: 142 MB on an M4 Max with a 50,000-line project (DevToolReviews); 180 MB against VS Code 650 MB and Cursor 690 MB (DevToolReviews, dated 2026-02-27, updated 2026-10-21, "Independent testing", no tool or column named); 222 MB with a folder open (TechNewsDaily). Zed publishes no idle figure. Discord's early build: about 290 MB (above). A gpui-kit 0.8 app peaked at 72.7 MB on Windows in Longbridge's own bench (see comparison.md). — https://www.devtoolreviews.com/reviews/zed-editor-review-2026, https://www.devtoolreviews.com/reviews/zed-vs-cursor-vs-vs-code-2026, https://technewsdaily.com/news/how-to-get-started-with-zeds-2026-multiplayer-code-editor/Partly. The direction holds (one native process, no Chromium); every specific number is a blog's unexplained measurement. Our number comes from the spike, measured with strobe
"GPUI reached 1.0"No. crates.io gpui stopped at 0.2.2 (2025-10-22); the README says "still pre-1.0. There will often be breaking changes between versions"; the kit pins weekly gpui-pre snapshots (0.3.12). What reached 1.0 is Zed the editor, on 2026-04-29. — https://raw.githubusercontent.com/zed-industries/zed/main/crates/gpui/README.md, https://zed.dev/blog/zed-1-0Does not hold. The editor's 1.0 is read as the framework's
"Zed stopped developing GPUI"On 2026-02-13 a Hacker News comment quoted a Zed maintainer's Discord post: "GPUI develoment is getting some major brakes put on it. We gotta focus on some business relevant work in 2026", with work "unrelated to Zed's own use case" pushed off; the same post says Nate, "former employee #1 at Zed", started gpui-ce. What "brakes" meant in practice: the tray PR by the kit's author closed "without a Zed use case" (2025-12), a shaders PR closed as not needed by the editor. What kept moving, because Zed needs it: wgpu on Linux (2026-02), the platform crate split, AccessKit (2026-05-27), weekly gpui-pre snapshots, 46 examples. — https://news.ycombinator.com/item?id=47003569, https://github.com/zed-industries/zed/pull/44047Partly. Zed stopped taking community features the editor does not need; GPUI itself moves at Zed's pace. The kit lives with this by pinning gpui-pre
"gpui-ce is the community's answer"gpui-ce/gpui-ce: created 2025-12-12, 1,573 stars, 160 forks, 31 open issues, Apache-2.0, pushed 2026-11-09 ("ai usage policy update (#393)"); crates.io gpui-ce 0.3.0 (2026-10-20), 25,465 downloads. README: "mostly API compatible with upstream, but that is changing!" Issue 280, "The State of CE in 2026" (2026-09-26): the ecosystem "is very fragmented", "We don't currently have any documentation or guides", "Ce currently merges changes from upstream gpui via an LLM"; a contributor names "gpui-kit's dominance" as a roadblock; locked and moved to a discussion on 2026-10-05. The kit's discussion 1856 asks for a feature flag to build on gpui-ce; gpui-tray offers one. — https://github.com/gpui-ce/gpui-ce, https://github.com/gpui-ce/gpui-ce/issues/280, https://github.com/longbridge/gpui-kit/discussions/1856Holds that it exists and is alive; small, undocumented, drifting from upstream. Not a path for us: the kit pins upstream
"GPUI runs on the web"Zed has a gpui_web crate; the kit's wasm build is "Showcase only" on nightly Rust; Native Shell targets the web through its own views. — see gpui-kit.md, native-shell.mdPartly. Not a shipping target for anyone's product
"GPUI Kit is three layers" (our plan.md)The docs count five (GPUI, the gpui-kit facade, gpui-base, gpui-component, gpui-shell); the repo's AGENTS.md says "three crates" plus the umbrella. — https://gpui-kit.com/docs, https://github.com/longbridge/gpui-kit/blob/main/AGENTS.mdPartly. Same stack, counted differently; gpui-kit.md has the picture

Who ships on GPUI, as of 2026-11-09

AppMakerWhat the evidence is
ZedZed Industries1.0 on 2026-04-29, 1.27.0 on 2026-11-04; 94,583 stars. — https://zed.dev/releases
Longbridge ProLongbridge (the kit's author's employer)Commercial trading desktop, first in the kit's App Stories. — https://gpui-kit.com/apps/
Cap, OpenLogi, disktree, Zedis, tty7, Loungy, Navop, Zeron, SonoraOpen-source teamsawesome-gpui entries with 1.5k to 25.4k stars (counts as the list shows them, not re-checked). — https://github.com/zed-industries/awesome-gpui
100 apps in the kit's App StoriesMixedDated 2026-09-05 to 2026-11-09; the vendor's own catalog. — https://gpui-kit.com/apps/
DiscordDiscordAnnounced 2026-11-04; nothing shipped (above)

Nothing in either list is a security product or a tray-first endpoint agent; the nearest shapes are a macOS menu-bar PR approver (hi5), a Windows shelf (Traydex) and a Linux network monitor (bandmeter).

Claims in our own documents that this reading retires

Said inRetireReplace with
plan.md Facts"gpui-component" as the product's nameGPUI Kit, renamed at 0.6.0 on 2026-09-03; gpui-component is now the styled layer's crate
plan.md Facts"three layers"Five in the docs; see above
General knowledge"Linux renders with blade"wgpu since 2026-02
General knowledge"the kit uses crates.io gpui"gpui-pre snapshots, pinned exactly
Hope"the kit has a tray" / "nsh ships a tray and an updater"Neither; nsh has them only on unreleased main
plan.md Facts"AccessKit" as "accessibility works"AccessKit is wired; no screen-reader run on record (gpui-kit.md)
plan.md Facts"strobe" as a released toolGit tags and npm packages, no GitHub release; three npm versions on 2026-11-09

Not verified

  • Vishnevskiy's X post itself (fetching X is not possible from here); everything about it comes from heise and dev.to.
  • The "Discord Phoenix" codename.
  • gpui-ce's founder: a quoted Discord post names Nate (Nate Butler, iamnbutler, who says in the same thread he no longer works at Zed); the repository names no maintainers.
  • The method behind every Zed idle-memory figure.

Sources (all read 2026-11-09)

https://dev.to/dishant0406/gpui-by-zed-the-electron-killer-that-never-left-home-209o · https://syntax.fm/show/821/is-tauri-the-electron-killer · https://www.heise.de/en/news/Discord-is-rebuilding-in-Rust-leaving-React-and-Electron-behind-11516113.html · https://dev.to/jamilxt/electron-vs-gpui-what-discords-290-mb-rust-rewrite-can-and-cannot-prove-34b6 · https://github.com/OscarArranz/rspotify · https://github.com/Letdown2491/riddim · https://github.com/crmne/spotifast · https://www.devtoolreviews.com/reviews/zed-editor-review-2026 · https://www.devtoolreviews.com/reviews/zed-vs-cursor-vs-vs-code-2026 · https://technewsdaily.com/news/how-to-get-started-with-zeds-2026-multiplayer-code-editor/ · https://raw.githubusercontent.com/zed-industries/zed/main/crates/gpui/README.md · https://zed.dev/blog/zed-1-0 · https://news.ycombinator.com/item?id=47003569 (dated through the HN API: 2026-02-13T15:08Z) · https://github.com/zed-industries/zed/pull/44047 · https://github.com/gpui-ce/gpui-ce · https://github.com/gpui-ce/gpui-ce/issues/280 · https://github.com/longbridge/gpui-kit/discussions/1856 · https://crates.io/api/v1/crates/gpui-ce · https://gpui-kit.com/apps/ · https://github.com/zed-industries/awesome-gpui · https://github.com/longbridge/gpui-kit/blob/main/AGENTS.md

Qpoint Brand Style Guide