Never lose what was said

Earkeep is a small, fast desktop app that records everything being said, transcribes it locally, and lets you mark which stretch of the day was a meeting after it already happened.

No bot · No cloud · No record button

14-day full trial, no account required.

Four ways Earkeep is different

A day of capture: an activity strip from 08:00 to 18:00 under a header reading "Recording since 07:12 · no button pressed", and below it the live transcript line listening to mic and system audio, with a Pause control at its end

Always-on, zero interaction

A quick huddle at someone's desk turns out to matter. You never hit record, and it was already being transcribed: Earkeep listens from the moment it is running, so nothing depends on you guessing in advance which conversation counts. Afterwards you go back and drag the span across the timeline. A voice-activity-detection pipeline transcribes only the parts where someone is actually speaking, so always-on doesn't mean always-transcribing.

Read how always-on works →

Everything inside one machine: microphone and system audio into memory, transcribed by Whisper on this device, out to a .jsonl file – with the network outside the boundary and nothing going to it

Audio never leaves your computer

Mic and system audio are processed only in memory and never written to disk, then transcribed on-device with Whisper or Parakeet.

No upload step, no API calls, and no account to create in the first place. The only network activity is an occasional license check.

Earkeep is private by design.

See how private-by-design works →

A ~/meetings folder of four .jsonl day files beside a terminal counting their lines, listing the day's meeting titles and pulling one sentence out of a file with grep

Plain files you own, on your own disk

Transcript saved on your own computer in human- and machine-readable format - you can open it in any editor, back it up however you like, and hand it to any tool you choose.

No database, no proprietary format, nothing to export: it will still be readable in ten years with no Earkeep involved.

Read the full plain-files story →

A meeting's own panel with an agent session on it: the question asked, the transcript it read, the decisions and open questions it answered, and a composer set to Codex

Your agents, right where your meetings are

Run the coding agents you already use directly inside Earkeep, against your meeting history: anything that speaks ACP connects, locally, with no configuration beyond picking it.

An MCP server exposes the same meetings to any MCP-capable tool, so an external client can read them too.

  • Claude Code
  • Codex
  • Cursor Agent
  • Gemini CLI
  • Hermes
  • Kimi CLI
  • OpenClaw
  • opencode
  • Pi
  • Qwen Code

See agents and MCP in Earkeep →

14-day full trial, no account required.

Pricing: one payment, no subscription

€39

once, for one device, not a subscription.

No tiers, no per-seat billing, no recurring bill - and no server on our side to charge you rent for. Try it free for 14 days with full functionality, no email and no account required. See full pricing and policy details →

Frequently asked questions

How does the 14-day trial work?

Download Earkeep and use it with full functionality for 14 days. No email address and no account are required to start the trial.

Does audio ever leave my computer?

No. Mic and system audio are processed only in memory and transcribed on-device with Whisper or Parakeet models. Nothing is uploaded to a server or an external API, and there is no cloud component to Earkeep's transcription.

Does Earkeep support my language?

Yes, across two models with different coverage. Whisper covers 99+ languages. Parakeet-TDT v3 covers 25 European languages, and tops the open ASR leaderboard on speed. See the full supported-language list per model →

How does licensing work across devices?

A license activates one device at a time. If you get a new device, deactivate the license on the old one and activate it on the new one; email [email protected] if you need help moving it.

Does Earkeep work offline?

Yes. Recording and transcription run entirely on-device and need no network connection. License validation uses an offline grace period, so Earkeep keeps working through normal periods without internet access instead of checking in on every launch.

How big is the model download?

From 75 MB to 3.1 GB, depends on your device processing power and transcript fidelity you want. A transcription model is downloaded once during setup and run locally afterward. See model details →

What macOS version do I need?

macOS 13 or later.

Does always-on capture drain my battery or slow down my device?

Earkeep uses a voice-activity-detection pipeline that only runs transcription when someone is actually speaking, filtering out silence and background noise. It uses the same local models as other transcription apps, so listening all day doesn't mean using more resources than a manual, start-and-stop transcriber would for the same amount of speech.

What is the refund policy?

Try Earkeep free for 14 days first, no email and no card required, so you can decide before you pay. If you do buy and it still doesn't work out, email [email protected] within 14 days of purchase for a full refund, no questions asked.

Stop losing what was said

14-day full trial, no account required.