Install
https://github.com/Natively-AI-assistant/natively-cluely-ai-assistant/releasesWho it is for
If what you need is an answer in the middle of a real conversation rather than a summary afterward, this belongs to the "live co-pilot" category. Its use cases are fairly specific: technical Q&A in job interviews, on-the-spot answers during pre-sales pitches and customer site visits, and note-taking in classes and meetings. It is not a background command-line program but a macOS desktop client, with an interface built around macOS conventions. Its core form is a layer of AI that floats above the current screen and can be questioned at any time (the project page describes it, in parallel terms, as a live coding assistant, screen overlay AI, and invisible AI assistant).
Its value also depends on what you already do: if you mostly communicate asynchronously through email and documents, the gain is limited; if you are in video meetings, pairing, or on site with customers every day, it can noticeably shorten the time you wait for answers and save you the effort of trying to remember later what you had asked.
Confirm two things before you choose. First, the licensing terms: the page describes itself as $0 for personal and non-commercial use, source available, and puts "no data leaks" forward as its public claim; if your use falls on the commercial side, it is not free and you need a separate licensing route. Second, the runtime platform: the available evidence covers only the macOS desktop client, and the page gives no information about other systems.
Your first attempt in 3 minutes
There is only one recommended path: the project's GitHub Releases page, which the page explicitly gives as the desktop download entry point.
https://github.com/Natively-AI-assistant/natively-cluely-ai-assistant/releases
Prerequisites: a Mac + a personal or non-commercial use case. After downloading the matching desktop package and confirming the system prompt, you need to grant screen recording and microphone capture; those two grants are themselves the "start capturing" action. Even though the project emphasizes that it runs locally, the first authorization should come only after you have reviewed the privacy implications.
Where it installs: the downloadable item on the page is a desktop build, and installing it gives you a local Natively client; the page does not spell out finer installation details (the package format, for example, or whether the app has to be dragged into the Applications folder), so just follow the system install prompts.
Trigger a first low-risk task: after launch, do not turn on invisible mode yet, and do not try it with a real meeting. Open an ordinary window that contains no sensitive information and type a read-only, harmless question in the overlay, for example "Summarize the window I currently have open in one sentence", and let the answer stay in the overlay. The success signal is that this text appears inside the overlay while your original window stays neither covered nor altered.
After that, enable the features that change behavior one at a time: microphone input (start recording), recording, and invisible mode. Confirm each one yourself before turning it on, and do not switch on all the recording and privacy-related toggles at once.
Use cases
Case one, real-time answers during interviews and pitches. The approach is to let the screen overlay read the current picture and then ask about the questions, code, or design mockups in it; the page describes this as live coding assistant and real-time coding help. It suits a candidate who writes code and needs a hint, or a pre-sales person who has to break a technical question down on the spot while the client waits. Evidence boundary: the page only gives a feature description and parallel keywords; it does not state the exact entry point, the keyboard shortcut combinations, or the interaction details of the question box, so you will have to confirm that yourself in the client.
Case two, drawing the system design directly with Visual Answers. When you are discussing system architecture, ask for Visual Answers and it will draw the system design out, so you can look at it while you talk instead of getting a block of text back. It fits system design interviews and solution-alignment sessions. Evidence boundary: the page explicitly says it draws system designs automatically, but does not explain which diagram types are supported, whether the result can be exported, or how the drawing can be modified.
What the two cases share is that neither requires you to prepare any material in advance: the overlay looks at the screen in front of you right now, and the answer lands in front of you as soon as you have asked.
Limits and boundaries
- Free and licensing: $0 covers personal and non-commercial use, and it is source available rather than under a standard open-source license, so commercial use is not included.
- Platform: there is evidence only for a macOS client; there is no information for Windows or Linux.
- Invisible mode: the page calls it the most advanced invisibility among the tools it has compared. Its purpose is to keep the overlay unnoticed, but used in someone else's interview or meeting it means having the other side answered without their knowledge; for a first attempt, use the visible overlay only.
- Capture and recording: a meeting assistant, note-taking, and recording all mean that the microphone and screen contents are read. Running locally lowers the risk of data leaving the machine, but it does not mean that nothing is captured; get consent before enabling it in someone else's meeting.
- Detail gaps: BYOK and local RAG appear in the project description, but the page does not explain the configuration entry points or the range of available models; if screen capture misbehaves on your first attempt, start with the permission entries in system settings.