Version 2 is an unreleased shell around chatgpt.com
The package calls itself version 2.0.0, while the default branch is v2-dev and its code opens https://chatgpt.com inside a Tauri webview. A small React interface supplies the title bar and a separate Ask panel. Native commands handle reload, navigation, always-on-top state, theme changes, and prompt submission into the hosted page. This is a desktop wrapper around ChatGPT's website, rather than a local model runner or a client built on OpenAI's API.
commit a6de9a8 contains 74 files and about 1,515 source lines in a 1.1 MB checkout. That is small enough to understand without weeks of archaeology. The Rust setup creates the window and remote webview; the TypeScript side draws controls; an injected ask.js bridge synchronizes the prompt bar with the page. The narrow size is attractive for study, though it also shows how much of the product remains owned by chatgpt.com and its changing interface.
The 25-second install stops at pnpm approval
Our fresh Node 22 sandbox ran the pnpm install for 25 seconds before it exited with code 1. The tail names ERR_PNPM_IGNORED_BUILDS, says the build script for esbuild 0.20.2 was ignored, and tells the user to run pnpm approve-builds to choose permitted dependencies. The same log says the lockfile passed its supply-chain policies. We cannot turn that into a successful setup result: the install step failed exactly where the supplied log says it failed.
The measurement block has no successful build or test result after that exit 1. It would be wrong to credit the TypeScript or Rust code with compiling merely because the repository is small. Approval is also a decision, not a command a CI job should accept blindly. A maintainer needs to review why esbuild needs its script, record the approved dependency policy, and then run a new controlled build. Our run did not perform that follow-up.
What happened when we ran it
Our sandbox used 3 CPUs, 8 GB of RAM, an unprivileged container, and no secrets. Install failed in 25 seconds at esbuild 0.20.2's ignored build script. The checkout scan found 0 CI workflow files, no Dockerfile, and no tests directory. Those are separate findings: pnpm's security policy caused the observed stop, while the missing repository signals mean there is no visible automated path here that demonstrates how maintainers approve, build, and test this branch.
The lab measured repository setup only. It did not launch a desktop session, authenticate to ChatGPT, exercise downloads, or inspect cross-platform packaging. No test count, build timing, vulnerability result, or runtime benchmark exists in the supplied block. The only defensible operational conclusion is that commit a6de9a8 did not install successfully under the stated clean conditions. Anyone evaluating a fork should first make that install reproducible, then test Linux, macOS, and Windows separately.
Three webviews keep the code small and the service dependency large
The v2 source assembles 3 webviews: the hosted ChatGPT page, a local title bar, and a local Ask input. Configuration stores theme, pinning, Ask mode, and a macOS title-bar preference under the user's config directory. Downloads are redirected to the operating system's download folder and opened through Tauri's shell plugin. No API credential is declared for the chat path because the app presents the website and leaves sign-in to that service.
This design inherits upstream behavior. The Ask bridge injects JavaScript into chatgpt.com and calls a page-side helper named ChatAsk; navigation controls also act on the remote view. A hosted markup or behavior change can therefore break local controls even when the 1.1 MB application itself has not changed. The Tauri configuration sets its content security policy to null and grants shell open and execute permissions. That is a reason to review the permission boundary before shipping a fork, not proof of an exploit.
Zero CI workflows leave interface changes unchecked
Our repository scan found 0 CI workflow files and no tests directory, while package.json defines development, build, preview, and Tauri commands without a test script. A remote-site wrapper needs regression checks because the owned code and the hosted page can change on different schedules. Even a modest smoke suite could verify window creation, URL restrictions, prompt handoff, navigation, and downloads. The current branch gives an adopter no visible automated evidence for those paths.
The manifest depends on Tauri 2 beta packages and declares Rust 1.77.1 in the desktop package. Beta-era dependencies are understandable in an experimental branch from 2024, but they raise the cost of reviving it in 2026. A new maintainer must reconcile the old JavaScript and Rust lockfiles with current platform tooling, then decide whether the remote permissions remain appropriate. Our failed 25-second install means that modernization work starts before the application can even be evaluated.
August 2024 was the last code activity despite 944 open items
GitHub reports a last push on August 29, 2024, while the newest release is v1.1.0 from August 3, 2023. The repository had 54,493 stars and 944 combined issues and pull requests when fetched. Users were still opening and updating issues in 2026, but that activity has not produced a newer source push on the default branch. Popularity explains why the project remains visible; it does not provide a maintenance path for a new deployment.
The README makes the status unusually plain. It calls Noi the successor, mentions OpenAI's official application, links back to v1.1.0 for the previous build, and says the author is looking for differentiating features for version 2.0. That last sentence is an aspiration, not a delivery plan. Start with Noi, NextChat, or Chatbox for an active client. Keep lencx/ChatGPT only if its 1,515 lines are useful as a readable example and you accept responsibility for every update.