Product notes

I made a Mac AI recorder: continuous recording and free local transcription

Xuesheng Guo

This article preserves the product version, test conditions and views at the time of writing. Features, pricing, data handling and download options may have changed; consult the current product page.

A small thing often happens while I'm working: someone comes over and gives me an instruction while I'm busy. I hear it, but by the time I return to the task, I can't remember exactly what they said.

The conversation may last only seconds and doesn't justify a meeting. I wanted a tool that could record for a long time and help me find the sentence later. So I made RecordAI; the version available at the time was for macOS.

Transcribe while recording, review any time

RecordAI has a continuous recording mode. It divides long recordings into segments and progressively finalizes completed segments while recording continues. To find a sentence, search the transcript and open the corresponding passage to replay it. Record a microphone in an in-person meeting, record microphone and system audio together online, or import existing audio.

Live text is displayed by default while recording. It is a draft; the final transcript and original audio remain the reference for checking names, numbers and decisions.

Local models, no transcription charge

Recording an hour or two and leaving a recorder running for long periods have different costs. Many tools make recording easy but charge a subscription or by duration for transcription. The more you use them, the more you need to watch your allowance.

At the time, RecordAI transcribed using a speech model on the Mac, with no per-minute charge, subscription or transcription-duration quota. The model is downloaded and verified on first use; after that, recording and transcription can work offline without an API key. Local models use disk space and computing power, but I find that more reassuring than constantly calculating transcription costs.

Local recordings and privacy

Some audio shouldn't go to the cloud. An internal meeting may contain an unpublished plan; a private conversation may not be something you want a third party to store. RecordAI performed recording, transcription and speaker analysis on the Mac, keeping audio and text locally. If you wanted to send results to DingTalk, Feishu or Notion, you previewed and explicitly confirmed the sync; text was sent, not the original audio.

For a privacy or compliance review, you can at least answer: where is the material stored, and when is it sent out? Local storage and user-initiated sync were the boundaries for that version.

Designed for agents, too

Once text is saved, I want an agent to help organize it. RecordAI publishes recording data to a local directory and offers CLI and MCP interfaces. A connected agent can read final transcripts and create conversation reports, to-dos and daily reports on request, or find a passing instruction and open the corresponding audio for verification.

Reports and to-dos are generated by the agent you connect. RecordAI handles long recordings, local transcription and access to the material. Live text is a draft; agents read finalized text by default and explicitly request drafts when needed.

Start on Mac

RecordAI also supports complete meeting recordings, generates a final transcript after stopping and attempts to distinguish speakers. Text can be edited and speaker labels renamed or merged. Transcripts can be exported as TXT, Markdown, JSON or original SRT, and individual audio segments can be exported. Speaker labels are inferred from voices; they do not automatically identify people.

RecordAI 0.15.1 required an Apple Silicon Mac running macOS 14 or later. A DMG or PKG was available through GitHub Releases. A speech model must be downloaded on first use. That package was not notarized by Apple, so macOS might require the user to allow it manually.

Next time someone gives you a quick instruction, finish what you're doing. When you return to it, the sentence can still be found.