Pluely v1: a stealth overlay assistant that ships as signed binaries, not source

What Pluely is for, and who it is not for

Pluely targets the moments where a browser tab costs you the room. The README names them directly: interviews, sales calls, standups, lectures, live debugging. The product is a translucent overlay that floats over whatever app you are already in, transcribes audio, and streams an answer without the other participants seeing it. It is not a meeting bot. Nothing joins the call as a participant, which is the architectural difference between Pluely and most recording assistants.

The audience is narrow and specific. Someone on a live call who needs a suggested answer before they speak. Someone watching a lecture who wants a running transcript with speaker labels. Someone debugging in a terminal who wants to drag-select a region of the screen and ask what is failing. The README's Ask mode example is exactly that: attach a screenshot of an error, ask what is breaking, get a streamed answer with follow-up suggestions.

It is the wrong tool if you need a shared team record. Sessions save locally as meetings with transcripts, but there is no server, no workspace, and no collaboration layer described anywhere in the repository. It is also wrong if you want to audit the code you run. The README states that v1 ships as signed binaries while the product stays free to use at its core, and the badge on the page reads Source: Closed (v1+). The repository you are looking at is the public shell around that decision, not the v1 application.

The overlay, the two modes, and where your data sits

The mechanism is a Tauri shell. The repository layout shows src/ for the React and TypeScript frontend and src-tauri/ for the Rust side, with Vite as the dev server and the Tauri CLI as the build entry point. That is what gets the installer down to the 9 to 16 MB range the README claims, against the Electron alternatives it compares itself to. The package.json confirms the stack: @tauri-apps/api, @tauri-apps/plugin-global-shortcut, @tauri-apps/plugin-autostart, @tauri-apps/plugin-http, React, Tailwind, and shadcn-style Radix primitives.

Two modes sit on top of that shell. Ask mode takes typed input, push-to-talk dictation, or a screen capture, including a drag-selected region, and streams Markdown answers. Listen mode runs live transcription of both mic and system audio, applies speaker labels and language selection, and fires automatic responses either when someone asks a question, after every pause, or only when you tap Suggest. Follow-up chips are generated from the conversation itself.

Storage is the part worth reading carefully. The README says chats, meetings, transcripts and files live in a local SQLite database on your machine, and that your own provider keys stay local. That is a real architectural commitment, and it is also the boundary: there is no sync, no shared history, and no server-side copy to recover if the local database is lost. Voice activity detection comes from @ricky0123/vad-react, which is what drives the pause-based triggers in Listen mode rather than a fixed timer.

Installing Pluely v1 and running a first Listen session

There is no build-from-source path documented for v1. The README points at pluely.com/download/macos, /linux and /windows, and the available formats are .dmg for macOS, .msi and .exe for Windows, and .deb, .rpm and .AppImage for Linux. The README also states that no account is needed to start and that updates ship automatically, so the first run does not involve a login wall.

If you want to work against the public repository instead, the package.json defines the scripts. This is the pre-v1 codebase, versioned 0.1.9, not the v1 binary:

npm install
npm run tauri dev

That starts Vite and the Tauri dev shell together. The other scripts are `npm run dev` for the frontend alone, `npm run build` for `tsc && vite build`, and `npm run tauri` to reach the Tauri CLI directly.

The first real use of the shipped app is Listen mode. The README describes the flow: hit Start, and Pluely transcribes your mic and system audio live, with speaker labels and language selection. Automatic responses are configured to fire when someone asks a question, after every pause, or only when you tap Suggest. If you want the conservative setting, choose the tap-to-suggest option, because the pause trigger depends on voice activity detection and will fire on any silence long enough to register.

For the free path, the README says you connect any LLM or speech-to-text provider through a curl template, or plug in the AI CLIs you already have: Claude Code, Gemini CLI, Codex, Qwen Code, Ollama. The curl template is the interesting piece, because it means the provider contract is defined by a request you write rather than a fixed integration list. Hosted models, described as 200+ on Pro plans, are the paid alternative and remove the key management entirely.

The closed-source v1 problem, and what the GPL-3.0 file still means

This is the part that decides whether Pluely fits your organisation. The README explains the reasoning in the project's own words: years of open Pluely code kept getting repackaged and sold as clones, and no license or complaint ever stopped it. The response was to ship v1 as signed binaries. The badge is explicit, Source: Closed (v1+).

The awkward detail is that the repository still carries a LICENSE file and package.json still declares "license": "GPL-3.0". That declaration belongs to the 0.1.9 codebase sitting in src/ and src-tauri/, which is what you would build with `npm run tauri dev`. It does not describe the app-v1.0.0 binary you download from pluely.com. Two artefacts, two different sets of obligations, and the README does not spell out which terms attach to the signed v1 installer. That is a question for whoever reviews software in your organisation, not something this article can settle.

A related gap: the README does not document rollback. Updates ship automatically, and there is no described mechanism for pinning an older release or reverting after a bad one. If you are deploying this on a managed machine, that is a real constraint. The release history shows app-v0.1.9 on 2026-01-14 and app-v1.0.0 on 2026-07-14, so the cadence between those two is roughly six months, but nothing in the repository describes how a downgrade would work.

Undetectability is a claim about capture, not a guarantee

The README's stealth claims are specific and testable in principle: excluded from screen capture, absent from recordings and screenshots, never steals focus from the app you are in, and the icon can disappear from the Dock or taskbar via the app-icon-stealth setting documented at docs.pluely.com. The mechanism that makes this plausible is the Tauri window configuration on the Rust side, and the fact that no bot joins the call at all, so there is no participant list entry to notice.

What the repository does not provide is a compatibility matrix. Screen capture exclusion on desktop platforms is implemented per-platform, and the README does not enumerate which capture tools, which OS versions, or which conferencing clients were verified. It also does not address mobile, browser-based capture, or a participant taking a photo of their own screen. Anyone treating "undetectable" as an absolute is reading past the claim's actual scope, which is the desktop capture path.

The honest framing: Pluely's stealth is a property of how the window is created and how the call is joined, and it is the strongest argument for choosing it over a browser extension. It is not a promise about every recording setup in existence, and the documentation does not make one.

How Pluely differs from Cluely and from the open-source Cluely clones

The repository's own description positions Pluely as the open source alternative to Cluely, and the topics list carries cluely-alternative. The practical difference today is the opposite of what that tag suggests. Cluely is a commercial product; Pluely v1 is also a commercial-style download, signed and closed, with a free tier and Pro plans carrying 200+ hosted models. What Pluely still offers that a pure SaaS does not is the bring-your-own-key path: any LLM or speech-to-text provider through a curl template, plus the AI CLIs you already have installed.

Against the older open-source Cluely-style projects, including Pluely's own 0.1.9 codebase, the difference is maintenance and polish rather than architecture. Those projects are source you can read and modify, and they are also the ones the README says kept getting repackaged and sold as clones. The v1 rebuild is the maintainer's answer to that, and it costs you the ability to read the code.

Against a general-purpose desktop assistant or a plain chat client, the difference is the overlay and the audio pipeline. A chat client does not transcribe your system audio, does not run voice activity detection to decide when to suggest, and does not hide itself from screen capture. If you do not need those three things, Pluely is more machinery than the problem requires, and a browser tab is genuinely cheaper.

Maintenance signals and what to verify before you commit

The last push to the repository was on 2026-07-14, the same date as the app-v1.0.0 release, and the repository is not archived. Before that, app-v0.1.9 landed on 2026-01-14 and app-v0.1.8 on 2025-11-14. That is a slow public cadence against a product the README says updates automatically, which is consistent with v1 development happening outside the public repository. Read the public commit history as the state of the pre-v1 codebase, not as the state of the app you would install.

Upgrade cost on the shipped app is close to zero by design, since updates ship automatically and there is no documented pinning mechanism. The cost you carry instead is the absence of rollback if an update breaks your setup, and the absence of a changelog in the repository beyond the release tags themselves.

On licence: package.json declares GPL-3.0 for the repository contents, and the README states v1 ships as closed binaries. Those two statements are not reconciled anywhere in the repository. If you are distributing Pluely internally, embedding it in another product, or running it under a policy that requires source availability, get the terms for the v1 binary in writing before you deploy. This article is not legal advice and the repository does not answer the question.

Editorial conclusion

Adopt Pluely if you want a local-first overlay for interviews, sales calls or lectures and you are comfortable running signed binaries from pluely.com rather than building from source. Do not adopt it if you need auditable code for a compliance review, or if you expect to run it on a platform outside macOS, Windows and Linux. Before you install, verify three things: which release the download page serves (the repository's latest app-v1.0.0 is dated 2026-07-14), whether the GPL-3.0 licence in package.json still governs the v1 binary you are about to run, and whether your own LLM or speech-to-text provider is reachable through the curl template, because the free path depends on your keys rather than Pluely's hosted models.

Frequently asked questions

Official sources

1. iamsrikanthnani/pluely on GitHub
2. License: GPL-3.0
3. Project website
4. README
5. Releases