Private by design

Two kinds of tools claim to transcribe your meetings, and both usually mean your audio travels somewhere else first:

• Cloud notetakers join your call as a bot and store the recording and transcript on their servers. Your meeting audio sits on infrastructure you don't control, governed by a privacy policy you didn't write.
• "Local" apps that still call out. Plenty of transcription apps run on your own machine but send the actual audio to a cloud API, OpenAI's Whisper API, Groq, or similar, to do the transcribing. The app is local; the transcription step isn't. Your recording still leaves the building.

Earkeep does neither. There is no bot, no upload step, and no third-party API in the transcription path.

The same problem has a second half that only shows up later: where the transcript ends up. Most notetakers keep your history on their servers, behind their app, their search, and their subscription. Stop paying and you typically lose access to months or years of meeting history, not just future recordings. Export is usually an afterthought, a button buried in settings that dumps a partial copy in a format the vendor's own app doesn't use internally, if it exists at all. Your 1:1 notes, the exact wording of a stakeholder commitment from eight months ago, the decision trail from a project that's now in dispute: all of it readable only on someone else's terms.

Audio lives in memory only

While Earkeep is running, it captures mic and system audio into memory for transcription, and that audio is never written to disk in any form, raw or intermediate. There's no audio file sitting in a cache folder, no temp .wav to find and delete. Once a chunk is transcribed, the audio behind it is gone; what remains is text.

Transcription runs on-device

Earkeep transcribes with Whisper or Parakeet models running locally, hardware-accelerated on Apple Silicon. This is the same class of model cloud services run on their servers, except it runs on your own hardware. No audio needs to travel anywhere for this step to happen.

No upload, no API, no server, no account

Putting the mechanism plainly: there is no upload step in Earkeep's transcription pipeline, no API call carrying audio to a third party, no server Earkeep talks to for transcription, and no account or sign-in required to use the app at all. If you disconnect from the internet, transcription keeps working, because it never depended on a connection in the first place.

No bot joins your calls

Earkeep doesn't add a visible participant to your video calls, and there's no separate app to authorize into your calendar just to "listen in." It captures system audio the same way it would capture any other audio playing on your computer, with no bot avatar, no third-party recording banner, and nothing for other meeting participants to notice or object to on their end.

License checks don't phone home on every launch

The one thing Earkeep talks to a server about is your license, and even that is deliberately minimal: validation uses an offline grace period, so the app doesn't need to contact anything every time you open it. It checks in periodically and keeps working through normal offline stretches in between. This has nothing to do with your meetings; it's the one non-audio network touchpoint in the whole app, and it's designed to be as infrequent as licensing allows.

Because there's no server, there's nothing holding your meeting history hostage. What Earkeep writes is one file per day:

• A daily transcript file, appended to continuously as Earkeep listens through the day, in JSONL (one JSON object per line): a plain, append-friendly, line-oriented format that survives partial reads, crashes, and being tailed or streamed like a log.
• The meetings you keep live in that same file, marked on the line where each one begins and the line where it ends. A saved meeting is a title and a stretch of the day, not a second copy of the transcript, so a correction shows up everywhere at once and deleting a meeting leaves every word that was said exactly where it was.

It is all just text. That means:

• Any text editor opens them. No proprietary viewer, no "open with Earkeep or nothing."
• You can search them with whatever you already use. A text editor's find, your file search, or the command line if that's your world: a folder of plain text files is searchable by ordinary tools, so finding a name, a phrase, or an old decision doesn't mean waiting on anyone's in-app search.
• Any backup tool already handles them. A scheduled backup, a synced folder, an offsite copy, or a private Git repository pushed to GitHub if you want a versioned history of your own record: whatever you already use to back up files backs up your meetings too, with no special integration required.
• You can sync them however you want. Since they're ordinary files, you choose the mechanism: a synced folder, a periodic copy, or nothing at all if you'd rather they stay untouched.
• Agents and scripts can read them directly. Point a coding agent or a script at the files and it can search, summarize, or cross-reference your meeting history the same way it reads any other text in a project. See agents and MCP for the in-app agents and MCP server built specifically for this.

There's also nothing to export, because nothing was ever locked in. Your transcripts are already plain files the moment Earkeep writes them.

• Always-on capture: how the transcript gets there in the first place, without a record button.
• Your agents, right where your meetings are: running agents against transcripts that never left your machine.