People search for a Screenpipe alternative for one of four reasons: the 5 to 10 GB it writes every month, the fact that a picture of the screen exists on disk, the license, or a price that starts at $21 a month for the full history. Sharpy answers the first two by design and the last two by being a different kind of product. This is where the two overlap and where they do not, from Screenpipe’s site and README on 18 September 2026 and from Sharpy’s own repository.
What Screenpipe records
Screenpipe calls itself “your computer history, open to any AI”. Its README says it captures the full accessibility tree with OCR as the fallback, microphone and system audio transcribed by Whisper Large-V3-Turbo on the machine or by Deepgram in the cloud, speaker identification, keyboard inputs and app switches. Capture is event-driven rather than a fixed frame rate. Everything lands in a local SQLite database with full-text search; screenshots are saved as JPEGs on disk.
The cost of that completeness is in the same README: 5 to 10 GB of disk per month, 8 GB of RAM recommended, 0.5 to 3 GB of RAM while running, 5 to 10 percent of CPU. It runs on macOS on Apple Silicon and Intel, on Windows 10 and 11, and on Linux if you build it. The code is on GitHub under the Screenpipe Commercial License, which allows personal, non-commercial use; it is source-available, not open source, and the company is backed by Y Combinator.
Pricing on 18 September 2026: Free at $0 (capture, search of recent work, sharing selected context with Claude and Codex, 10 AI credits a month), Basic at $21 a month or $250 a year (full searchable history, reusable context over MCP, unlimited scheduled workflows, 150 credits), Business at $42 a seat a month (sync across devices, team workspace, 400 credits), Enterprise on request. Students get 50 percent off the first month.
Pixels or words
That is the whole difference, and every other row in the table follows from it.
Screenpipe keeps pictures of your screen and the sound in the room, then extracts text from them. You get the closest thing to a recording of your day, and you pay for it in disk, memory and the knowledge that the frames exist.
Update, 26 October 2026: Sharpy has since become an AI to-do list for Mac. It still reads the text on your screen the same way and keeps it on your Mac, and what it does with that text now is write your list of tasks and cross them off. The comparison below is updated to what Sharpy does today.
Sharpy keeps the words. It reads the text of the front window through macOS Accessibility several times a minute, runs it through a privacy gate, and keeps it on your Mac. No video, no screenshots, no keystrokes, no audio file. Card numbers, keys, phone numbers and codes are masked before anything is stored; passwords, banking, logins and private windows are never saved. From that text it writes your to-do list, each task linked to the moment it came from.
On one founder’s Mac the September 2026 history is 24,000 snapshots in a 192 MB database. Screenpipe’s own estimate for the same month is 5 to 10 GB.
Side by side
| Sharpy | Screenpipe | |
|---|---|---|
| What it records | Text of the front window via Accessibility, several times a minute. Calls transcribed by a cloud speech model; only the text is kept. | Accessibility tree, screen frames with OCR fallback, microphone and system audio transcribed, speaker identification, keyboard inputs, app switches. |
| What sits on disk | One SQLite file of text. 24,000 snapshots in 192 MB on one Mac. | SQLite with full-text search plus JPEG screenshots. 5 to 10 GB a month per the README. |
| Resources | Menu-bar app; text capture only, no video or audio pipeline. | 8 GB RAM recommended, 0.5 to 3 GB used, 5 to 10 percent CPU per the README. |
| Where the data lives | On your Mac. The relevant text goes to cloud models to write the list and answer. | Local by default. Deepgram transcription and the Business tier’s cross-device sync go through the cloud. |
| What you do with it | A to-do list written from what you saw and the apps you connect, closed when the work is done; ask about your day with a citation; tap the right Option to draft a reply in your tone. | Search; “pipes” such as meeting notes, CRM update, time tracker, SOP generator, contact dossier; scheduled workflows. |
| AI tools | MCP server for Claude Code, Codex, Cursor, ChatGPT and any MCP client: the list (list_work, add_work_item, update_work_item) and memory search (search_memory, get_recent_activity). | MCP server for Claude Desktop, Cursor, VS Code with Cline or Continue; local REST API. |
| Platforms | macOS 14 or later, Apple silicon. Windows on the waitlist. | macOS (Apple Silicon and Intel), Windows 10 and 11, Linux from source. |
| Source | Not open source. The MCP tools are documented at getsharpy.com/mcp. | Source-available on GitHub, Screenpipe Commercial License, personal non-commercial use. 21.6k stars. |
| Price | Free in early access. | Free $0; Basic $21 a month or $250 a year; Business $42 a seat a month; Enterprise on request. |
The Screenpipe column is from screenpipe.com, its pricing page and the GitHub README on 18 September 2026. The Sharpy column is from the app’s repository. Check both before you decide; these products change monthly.
When Screenpipe is the right choice
You want the recording. If you need to see what a window looked like, not just what it said, Screenpipe keeps the frame and Sharpy does not. If you need transcripts of every meeting with speakers separated, Screenpipe records the room; Sharpy transcribes a call it detects on screen and keeps the text, nothing more.
You are on Windows or Linux today. Sharpy is Mac first, and Windows is a waitlist, not a download.
You want to read the code. Screenpipe’s is on GitHub. It is not open source in the OSI sense, but you can inspect what runs.
When Sharpy is
You liked the idea and not the 5 to 10 GB. Text of what you saw is a few hundred megabytes a month, and it is searchable the same way.
You do not want a picture of your screen on disk. Sharpy never takes one for your history. The privacy gate masks card numbers, keys, phone numbers and codes before anything is stored and skips passwords, banking, logins and private windows entirely.
You want the memory to turn into a list. Sharpy reads what you were asked and what you promised, in Slack, mail, calls and any app on screen, and keeps one to-do list of it. A sent reply or a merged pull request crosses a task off; a task that went stale asks to be dropped. Tap the right Option in any text field and Sharpy drafts the reply in your tone. Screenpipe’s pipes run workflows on a schedule; Sharpy keeps the list current as you work.
You use Claude Code or Cursor. Both products speak MCP. Sharpy’s gives the agent your to-do list as well as search over what you saw, and its memory rows cite the snapshot, so the agent can show you where a fact came from. See Claude Code memory: what it keeps, and what it can’t see for the setup.
The limits, stated plainly: Sharpy needs an account, every model call goes through the Sharpy gateway, and providers may keep prompts for up to 30 days for abuse monitoring. Your history stays on the Mac; the context for a question does travel to a model.
Moving from Screenpipe
Sharpy does not import a Screenpipe database today. Its SQLite is yours to keep. If you run both for a week, they do not conflict: Sharpy reads the accessibility text of the front window, Screenpipe records the screen, and each keeps its own file.
Sharpy is free in early access on Macs with Apple silicon: join the waitlist, and access goes out by email in waves. For the wider field, including Goldfish and HeyClicky, read Rewind AI in 2026: what happened, and what replaced it.
