Keen Bean drafts specs live from Mac meeting audio

Why it matters

Keen Bean pushes meeting AI from recall into production. Its test is whether live drafts save enough follow-up work to justify another subscription and cloud processing.

Michael Jones, an Australian independent software maker, built Keen Bean around a pointed pitch: meeting software should begin the document people need after the call, rather than leave them with another transcript to process.

Keen Bean listens to both sides of a meeting through a Mac's audio, without entering Zoom, Google Meet or another call as a visible participant. As people talk, Keen Bean says it identifies decisions, tasks, risks and blockers, then starts drafting artifacts such as a specification, diagram or rough interface mock-up.

Jones operates Keen Bean through Savvy Monkey, his sole-trader business in New South Wales, according to Keen Bean's terms of use. That small-business structure shows up throughout the product. Keen Bean runs on the user's Mac, keeps its output in local files and targets the person who conducts a meeting and then has to produce the resulting work. Jones's product thesis, stated on Keen Bean's homepage, is blunt: "a transcript is useless to you, because you were never going to re-read it."

Keen Bean is deliberately narrower than the meeting platforms selling shared workspaces, administration controls and company-wide archives. Jones is building for consultants, solution architects, founders and other independent operators whose calls are expected to produce a proposal, technical plan or product document.

The meeting becomes a work session

During a call, Keen Bean presents a newest-first feed that can sit beside the meeting window. Users can type questions silently or give the app spoken instructions such as "hey Bean, mock that up." The software merges duplicate cards, puts blockers first and preserves dismissed material behind a "show all" control, according to Keen Bean's product page.

The promised output goes beyond summaries. A discussion about an import screen, for example, can produce a rough HTML prototype while the meeting is still running. Keen Bean characterizes those outputs as editable starting points.

That boundary is central to Jones's bet. Keen Bean does not have to finish a client-ready specification to save time. It has to capture enough structure, decisions and early drafting that the user starts the evening with material to edit instead of an empty document.

Users can export the summary, transcript, cards and generated drafts directly to an Obsidian vault or another project folder. Keen Bean also offers search across previous meetings, including questions about ownership, earlier decisions and objections raised by participants.

The files remain readable outside Keen Bean, including after cancellation. That portability gives Jones a practical answer to a familiar risk with small subscription products: customers retain the work even if they stop paying or the software eventually stops receiving updates.

Botless capture is the entry ticket

Capturing desktop audio without placing a bot in the meeting is already an established product choice. Granola markets an AI notepad that uses computer audio without joining the call, while Read AI offers botless desktop capture on Windows and macOS.

Keen Bean's positioning rests on what happens after capture. Jones is betting that live artifact creation and local file ownership can provide a sharper reason to switch than another variation on transcripts, summaries and searchable meeting memory.

That focus also sets a boundary around the market Keen Bean can serve. The app supports macOS 13 and later on Apple Silicon and Intel Macs. Keen Bean explicitly excludes Windows and Linux users, and it is not designed for companies seeking a shared workspace, SSO or centralized administration.

For an owner-operated product, the narrow lane is coherent. Jones can build around the workflow of one person who owns both the conversation and the deliverable, rather than matching the collaboration and compliance checklists of larger meeting-software vendors.

Local files do not mean local AI

Keen Bean keeps meeting files on the Mac and says its own servers do not proxy or store the content. The processing path still includes outside AI providers.

According to Keen Bean's privacy page, meeting audio travels directly from the user's Mac to Deepgram for transcription. Transcript text then goes to OpenRouter for subscribers, or to another model provider selected by users supplying their own configuration. Keen Bean says meeting content does not pass through Savvy Monkey's servers.

Customers working under strict client confidentiality rules will still need to evaluate that processing path, even though Keen Bean itself says it does not receive the audio or transcript.

The invisible capture also does not remove the need for consent. Keen Bean asks users to confirm once per meeting that participants know transcription is happening, and its terms of use place responsibility for complying with recording laws on the user. The app can stay out of the participant list; the person running the meeting still has to disclose the recording.

A subscription built around AI costs

Keen Bean offers a 14-day trial and two paid plans. Seed costs $19 per month, while Sprout costs $39 per month and includes twice the AI allowance. Annual billing includes two months free.

Both subscriptions cover the Mac app, transcription and model usage through capped provider keys. A meter shows the remaining allowance in hours. When a user exhausts it, Keen Bean continues recording and transcribing, while artifact generation pauses until the allowance resets or the user buys a top-up.

The arrangement saves users from opening provider accounts or reconciling separate API bills. It also gives Jones a direct constraint: the subscription price has to cover variable transcription and model expenses. Keen Bean does not offer a permanent free tier, consistent with that cost structure.

Jones is entering a crowded meeting-assistant category with a product built around a specific moment of frustration: the call is over, the decisions were made, and the actual document still does not exist. Keen Bean's case is that the useful meeting record should be the first version of that document.