what this is for mac users

macOS already ships voice memos and quicktime, but both have failure modes that bring people to a browser tool: voice memos sync to icloud whether you want it to or not, and quicktime's "new audio recording" is buried two menus deep and exports a .m4a that not every transcription service accepts. this page records straight from the mic in your browser tab, gives you a downloadable file in webm, mp3, or wav, and never sends a byte to a server.

it works in every modern mac browser: safari 14.1+, chrome, arc, firefox, edge. the codec the browser produces differs (chrome and firefox give you webm/opus; safari gives you mp4/aac), but the page handles that transparently and the download names itself with the right extension.

microphone permissions on macOS

macOS gates microphone access twice — once at the browser level, once at the system level. if recording fails silently, it's almost always one of the two. the steps below assume macOS Sonoma or later; older versions have similar paths under System Preferences.

browser-level permission. when you click "start recording," your browser shows a prompt asking for microphone access. allow it. if you missed it or accidentally clicked block, look in the address bar for a microphone or "blocked" icon and click it to undo.
system-level permission (the one that catches people). open system settings → privacy & security → microphone and confirm your browser is in the list and toggled on. if a browser is missing entirely, quitting and reopening the browser usually triggers macOS to add it the next time it requests the mic.
per-site setting in safari. safari → settings → websites → microphone shows a row per site you've granted or denied access to. if the site is there set to "deny," change it to "allow" and reload.

airpods, bluetooth headsets, and the latency tax

the second-most-common source of bad mac recordings is bluetooth audio. when airpods or any bluetooth headset is connected and selected as the input device, macOS switches the entire system into hands-free profile (HFP / SCO), which drops audio quality dramatically — narrowband mono at ~8khz. fine for a phone call, miserable for an interview recording. the recording will sound like it was captured through a tin can.

two ways to handle it. either (a) record using your mac's built-in microphone, then listen back through the airpods — switch the input source in system settings → sound → input to "macbook pro microphone" while leaving output on airpods; or (b) use a wired headset or usb microphone, which both stay in full-bandwidth mode. for serious interview work, a $30 usb mic on the desk beats $250 airpods every time.

which input device the browser picks

chrome, arc, and firefox don't expose a per-page mic picker — they grab whatever is set as the system default input. switch defaults via system settings → sound → input, or click the volume icon in the menu bar with the option key held to get a quick switcher. safari respects the site-permission's last-used input but doesn't let you change it mid-recording.

test before the interview starts. record ten seconds, play it back, confirm levels look reasonable. an interview where you discover after the fact that the wrong mic was selected is unrecoverable.

file formats and what to do with them

three download options:

webm (chrome, arc, firefox capture this natively) — smallest file, ~720KB per minute at 128kbps. works as input to every modern transcription service we've tested, including audiohighlight. opens in vlc and quicktime via the perian/etc plugins, though the macOS native quicktime player won't recognize it without help.
mp4/m4a (safari capture, no conversion needed) — universal mac compatibility. opens in voice memos, garageband, logic, and quicktime out of the box.
mp3 — convert to this if you're uploading to an older transcription tool, a podcast host, or anywhere that specifies mp3 only. the conversion runs locally via ffmpeg.wasm in the same tab; nothing uploads. takes a few seconds per minute of audio.
wav — uncompressed PCM, ~10MB per minute. for archival, audio engineering, or broadcast ingest. lossless, no quality tradeoff in subsequent processing.

nothing uploads — for real

most "online voice recorder" tools quietly stream your audio to a server while you record. some advertise this as a feature ("save to the cloud"); some don't mention it but do it anyway. for a voice memo it's usually fine. for an interview recording, an attorney-client conversation, chart notes dictated between patients, source-protected reporting, or anything you'd be uncomfortable having a saas company subpoenaed for — it's not.

this tool runs the entire capture path in the browser tab. open the network panel in safari's web inspector or chrome's devtools while you record; the only request is the page itself loading. when you close the tab, the recording goes with it unless you've already downloaded it. for the structural argument and how to verify it from your own inspector, see this post.

what to do with the recording

transcribe it. audiohighlight's transcription workspace is the obvious follow-on. drop the file in for a transcript with bulk speaker-fix and click-word-to-replay-audio. join the beta waitlist.
trim it. run the recording through the audio trimmer to cut the silence at the front, the goodbye at the back, and any pause-for-the-server moments in the middle. also runs locally, no upload.
convert format. if the recording is webm and you need mp3/aac/flac for a specific downstream tool, the same-page conversion will cover most cases; for unusual targets, use the dedicated audio converter.
publish. webm and mp3 both upload cleanly to most podcast hosts, social platforms, and content management systems. mp4/m4a works for apple-ecosystem destinations.

limits

single-channel only. the browser's MediaRecorder API captures one input. for a two-mic interview where each speaker should be on a separate track, you need a dedicated DAW (logic, reaper, audacity). this is a different category of tool.
no system-audio capture. the page records from the microphone, not from "what's playing on the mac." for that, use loopback, blackhole, or screen-share-with-audio. those route the system mix into a virtual input device that we can then capture, but the setup is non-trivial.
memory pressure on long recordings. the recording lives in browser memory until you click download. on a 16GB mac you'll go past 90 minutes before anything strains; on 8GB machines the practical ceiling is closer to 60. for half-day depositions, record in chunks.