Two very different kinds of meeting tool
Most products described as AI meeting assistants are bots. They join the call as a visible participant, record it, and email you a summary afterwards. That model works well for recurring internal meetings where everyone expects it, and badly everywhere else — the bot appears in the participant list, it usually needs calendar and meeting-platform permissions, and it gives you nothing at all while the meeting is happening.
The other model runs locally. It captures the audio your computer is already playing, transcribes it on your machine's behalf, and helps you during the conversation. No bot joins, no calendar integration, no per-platform setup. JobSeeker AI is this second kind.
How it works with any platform
Because it captures at the operating-system level rather than integrating with a specific service, platform support is not really a feature list — it works with whatever is producing sound.
System audio is the output from your speakers: everyone else on the call, whether that is Zoom, Microsoft Teams, Google Meet, Slack huddles, Discord, a webinar or a recorded video.
Microphone audio is you.
Both are captured simultaneously and merged into one running transcript, so the record reflects the actual back-and-forth rather than one side of it.
Nothing needs to be installed inside Zoom or Teams, and no admin has to approve an app for your workspace — a genuine advantage if you work somewhere that locks down third-party meeting integrations.
What it gives you mid-meeting
The live transcript is the foundation, but the useful part is what sits on top of it.
You missed something. The transcript is right there — scroll back rather than asking someone to repeat themselves.
A question lands on you. The assistant detects it and drafts an answer using the documents you have uploaded.
Someone shares a diagram or spreadsheet. Capture a screenshot and the app runs it through a vision model, so you can ask about something on screen rather than squinting at it.
A term comes up you do not know. Ask in the side panel without interrupting the call.
You need to take notes while listening. On Magic and lifetime plans, Whisper Dictation types what you say straight at your system cursor, into whatever app you have focused.
Using your own documents as context
A meeting assistant that only knows what was said in the last ten minutes is not much help when a client asks about a commitment from a previous quarter. Upload the relevant material and answers get drawn from it.
The knowledge base accepts PDF, DOCX, PPTX, TXT, CSV, HTML, Markdown, JSON and SQL, along with source files in more than 40 programming languages. Typical things worth loading before a call: the project brief, the previous meeting's notes, the statement of work, the API documentation, the pricing sheet.
Indexing happens locally and the index stays on your machine. Only the question you ask and the passages relevant to it are sent to an AI provider.
Keeping the assistant off a shared screen
The obvious problem with an on-screen assistant is the moment you share your screen. JobSeeker AI has a screen privacy mode that marks its own window as excluded from capture, so it does not appear in what you are broadcasting while remaining visible on your physical display.
This is a Windows display-capture feature rather than anything clever, and it behaves accordingly: it applies to the app's own window, not to your whole desktop, and it is there so your notes and drafts are not accidentally projected to a room of people. Treat it as protection against an embarrassing slip, and be straightforward with colleagues about using an assistant — most people care far less than you would expect.
What a meeting costs
| Action | Credits | Notes |
|---|---|---|
| Live transcription | 2 | Per billed interval while the mic is open |
| AI answer | 1 | Per question, typed or detected |
| Screenshot analysis | 4 | Sends an image to a vision model |
Transcription is the recurring cost in a long meeting, so it is worth turning capture off during stretches you do not need recorded. New accounts start with 30 free credits and purchases are one-time credit packs, not a subscription.
Being sensible about consent
Recording and transcription law varies by country and by state, and in several places every participant has to consent. A local tool that does not announce itself makes it easy to forget that, so the responsibility sits with you rather than with the software.
The practical version: say at the start that you are taking AI-assisted notes. It takes four seconds, it is almost always met with indifference or interest, and it removes the entire problem. For anything covered by a confidentiality agreement, check whether sending content to third-party AI providers is permitted before you upload it.
Run it on your next call
Thirty free credits cover a real meeting end to end — live transcript, a few questions against your own documents, and a screenshot or two. No bot in the participant list, no workspace admin approval.
