---
title: A Screenpipe alternative that doesn't record all day
description: Screenpipe records your screen and audio around the clock. Minne remembers your work from the text of your windows instead — free, open source, plain markdown on your Mac.
---


[Screenpipe](https://screenpipe.com) is the best-known answer to "how do I
give an AI a memory of my computer": it records your screen continuously,
runs text recognition over the frames, transcribes your microphone and system
audio, and keeps all of it on your machine for you and your AI agents to
search. If that is what you want, it is a serious tool and this page will not
talk you out of it.

Minne is the other answer. It never records the screen. It reads the *text* of
the window you are working in through the macOS Accessibility APIs — the
interface VoiceOver uses — and distils it into a wiki of plain markdown notes
in a folder on your disk. It is free, MIT-licensed, and it does one thing a
recorder does not: tap right-Option in any text field and it writes the reply
or the summary from what it remembers.

> A recorder keeps everything you saw and lets you search it later. A memory
> keeps what you did, as notes you can read. They are different tools, and
> the choice is about how much of your day you want on disk.

## What Screenpipe does well

From Screenpipe's own site and pricing page, checked at the date in the
footer:

- **Total capture.** Screen frames with OCR, microphone and system audio
  transcribed, plus accessibility context — three channels, all day. If it
  was on your screen or said in your meeting, it is searchable afterwards.
- **Local by default.** Data stays on the device with cloud upload off, and
  there are controls to exclude sensitive apps, windows and URLs before
  storage, with redaction of fields like passwords and card numbers.
- **Built for agents.** The capture is exposed to AI agents and pipelines,
  which is the product's centre of gravity — it is a data layer for AI as
  much as an app for people.
- **Three platforms.** macOS, Windows and Linux. Minne is Mac only.
- **A free tier.** Free with limited searchable history; Basic is $21 a month
  or $250 a year for the full history; Business is $42 per seat a month. The
  code is source-available rather than open source.

## What changes when nothing is recorded

Minne asks for one permission — Accessibility — and never for Screen
Recording, so the purple recording indicator never appears because of it.
What it reads is text, as text, from the focused window only: the field you
are in, the visible text around it, the title, the URL. Then it distils.

That single choice cascades:

- **Small.** A day of text captures is kilobytes, not a video archive. There
  is no retention setting to agonise over because there is nothing large
  to retain.
- **Readable.** The memory is markdown — one page per person, project and
  topic in `~/Minne/wiki/`, raw captures under `~/Minne/sources/`. Open it in
  Obsidian, grep it, edit it, delete a page you don't like.
- **Handled before it is stored.** Password fields are skipped before a
  character is read; card numbers, IBANs and personal ID numbers are masked
  before anything touches disk. A frame, by contrast, contains whatever was
  on screen. [What leaves your Mac](https://docs.minne.sh/privacy) lists the
  rest.
- **Your model, or none of theirs.** Bring an Anthropic or OpenAI key, sign in
  with a subscription, or point Minne at [Ollama on localhost](/local-ai-writing)
  — and then nothing leaves the machine at all.

And the reason to remember anything: Minne writes from it. Tap right-Option
in Mail, Slack or a blank doc and the [email reply](/email-replies) or
[status update](/status-updates) arrives already knowing the thread and the
history. Search — ⌥Space, ask *what did we agree with Ingrid?* — is there too,
with citations to the captures it read.

## Side by side

| | Minne | Screenpipe |
| --- | --- | --- |
| What is captured | Text of the focused window, via Accessibility | Screen frames + OCR, mic and system audio, accessibility context |
| Audio and meetings | No | Yes, transcribed |
| Screen Recording permission | Never requested | Required for frame capture |
| Stored as | Plain markdown files | Screenpipe's local database |
| Writes into your apps | Yes — tap right-Option in any text field | Not the product's focus |
| Model | Your choice: API key, subscription, or Ollama locally | Per its site, with monthly AI credits per tier |
| Platforms | macOS 14+ | macOS, Windows, Linux |
| Price | Free | Free tier; Basic $21/mo or $250/yr |
| License | MIT, [code public](https://github.com/mandgie/minne) | Source-available |

## Choose Screenpipe if…

- Meetings are the thing you want remembered. Minne does not touch audio.
- You need to see what a window *looked* like, or you work in tools that are
  essentially pictures — design canvases, video, dashboards.
- You are building agents over a rich stream of everything, not a person
  looking for a reply.
- You need Windows or Linux.

## Choose Minne if…

- The words are enough — threads, decisions, people, what was agreed — and
  you would rather not keep a recording of everything you saw.
- You want the memory as files you can open and take with you.
- You want to pick the model, including a fully local one, and pay nobody
  for the app.
- You want it to *write* for you, not only to be searchable.

Minne is early and says so. The same honesty applies in the other direction:
the case for reading text instead of pixels is made in full at
[your Mac's AI memory doesn't need a screen recorder](/ai-without-screen-recording),
and the closest tools that share Minne's approach are compared at
[Minne and OpenHistory](/vs/openhistory) and [Minne and Goldfish](/vs/goldfish).
If you arrived here because Rewind shut down, [that page](/rewind-alternative)
is the one you want.