ScreenPipe vs Everpaper: honest 2026 comparison

Everpaper is our product, so read this with that bias in mind. ScreenPipe is one of the most interesting local-first capture projects around, and we think it deserves a fair fight. This comparison is for people deciding between a hackable recording framework and a finished private daybook.

Quick verdict#

Choose ScreenPipe if…

• You want to build your own tooling on top of raw screen and audio capture.
• You live in the terminal and prefer a CLI over a GUI.
• You need Windows or Linux support, or you're on an Intel Mac.
• You're fine assembling your own workflow from pipes, SQLite, and an HTTP API.

Choose Everpaper if…

• You want a finished product: day journal, meeting recording, and a memory you can ask, working out of the box.
• You value polish: ⌥Space Concordance, end-of-day review, Calendar Auto-Start.
• You want on-device transcription with speaker labels (Parakeet and Nemotron, with MOSS for Detailed meetings) within a guided meeting workflow.
• You're on Apple Silicon and macOS 15 or later.

At a glance#

How we compared#

We compared the documented capture models, transcription options, meeting workflows, AI architecture, platforms, licenses, and pricing. ScreenPipe's official docs, site, license, and GitHub were the sources for its behavior; Everpaper's shipped docs were the source for its claims.

Head-to-head#

Capture model#

Verdict: ScreenPipe for raw throughput, Everpaper for curated signal.

ScreenPipe records everything. Screenshots plus accessibility-tree text on meaningful actions, microphone, system audio. Default capture rate is 1 FPS, and you can lower it if your machine runs hot. The result is a dense, queryable record of your entire day stored in a local SQLite file. If you want to grep your life, this is the tool.

Everpaper is more selective. The Day Journal takes a full-screen screenshot every 2 minutes by default (30 s to 10 min), or keeps a screenshot-free activity timeline. Meetings are recorded with separate microphone and system-audio tracks. Memory (also called Second Brain) builds a knowledge graph across days and reads Ready after 5 days, 300 frames, 60 distinct things, and 3 recurring threads.

ScreenPipe gives you everything and lets you decide what's signal. Everpaper decides some of that for you, which saves time if you trust the defaults.

Transcription and diarization#

Verdict: Everpaper for meetings, ScreenPipe for flexibility.

Everpaper's speech models run on your Mac after a one-time download. For meetings, Parakeet (parakeet-tdt-0.6b-v3) writes the words and Nemotron labels speakers live (the default, Everyday). For harder calls, Detailed re-listens with MOSS (~2 GB), which transcribes and separates speakers in one pass after you stop. The result is a speaker-labeled transcript, live during the call and saved with the meeting; a meeting's Speakers tab (⌘4) can hide your voice or any speaker from that meeting's transcript view, while the audio and exports keep everyone.

ScreenPipe uses Whisper locally, with sizes from whisper-tiny up to whisper-large-v3-turbo. Cloud transcription via Deepgram is also available. The official FAQ reviewed for this comparison did not document speaker identification or diarization.

AI features and architecture#

Verdict: ScreenPipe for composability, Everpaper for coherence.

ScreenPipe's AI story is pipes. A pipe is a pipe.md prompt plus a schedule. When a meeting ends, a pipe can summarize the transcript, extract action items, and write an Obsidian note. Pipes run on local models via Ollama or on cloud providers (Claude, ChatGPT, Gemini). The MCP server lets Claude Desktop, Cursor, Codex, and VS Code query your capture history. If you want to build an agent that watches your screen and does something about it, ScreenPipe is the substrate.

Everpaper's AI is productized. Ask your memory… (⇧⌘A) streams prose answers from matching days and frames, with sources you can click back to the original screen. The Reflection writes up to three observations from real captured moments, each pointing back to evidence. Both work on-device with Everpaper Native (macOS 27) or a local oMLX, LM Studio or Ollama server, or with a cloud provider after you opt in. The difference: Everpaper's AI features are designed to work together, not to be recombined.

Search and recall#

Verdict: ScreenPipe for API access, Everpaper for immediate utility.

ScreenPipe exposes a local HTTP API at localhost:3030. You can query frames, OCR text, audio transcripts, and pipe outputs from any language. The search UI is functional, but the API is the point.

Everpaper's Index (⌘K) jumps to any day, meeting, belief, saved word or phrase, setting or command. The Concordance (⌥Space) floats a search/ask panel over any app, with ⌘1–4 to scope results and ⌘Return to continue in Memory. Without an AI provider, recall still returns cards: phone numbers, emails, links, files with time-and-place stamps. ScreenPipe's API is more powerful if you write code. Everpaper's recall is more useful if you don't.

Privacy and data handling#

Verdict: Both are local-first; Everpaper's redaction is more aggressive.

ScreenPipe stores data in ~/.screenpipe/db.sqlite and data/. Capture, OCR, local Whisper, and search work fully offline. Cloud features (Deepgram transcription, Claude/ChatGPT/Gemini chat, archive, team sync) require network. The homepage claims password managers are excluded and cards and keys are redacted before storage. A default-on PII removal flag scrubs email, phone, and card numbers.

Everpaper keeps everything on your Mac. Capture, OCR, transcription, knowledge graph, beliefs, and search run offline once the speech model has downloaded. Redaction runs twice: once at capture time, again before every outbound AI call. It scrubs private keys, JWTs, API keys, bearer tokens, payment card numbers, SSNs, and OTPs into [REDACTED:api_key]-style placeholders. Local AI sees real names and emails; secrets are scrubbed for every provider. External agents via MCP are off by default. Turned on, they get people, files, links, and connections with secrets removed (names and email addresses are kept), never screenshots; raw screen text needs a second switch. What an agent reads then goes wherever that agent sends it.

Both are honest local-first tools. Everpaper's double-pass redaction and explicit provider gating are more conservative.

Platform support#

Verdict: ScreenPipe, by a wide margin.

ScreenPipe runs on macOS (Apple Silicon and Intel), Windows 10/11, and Linux. The Linux CLI has full support; the desktop app is partial and best on X11, with Wayland varying by compositor. If you're on an Intel Mac, a Windows machine, or a Linux box, ScreenPipe is the only option here.

Everpaper requires macOS 15 or later and Apple Silicon. Its speech and AI models are built for Apple silicon, so Intel Macs are out. There is no Windows, Linux, iOS, Android, or web version. If you're not on an Apple Silicon Mac, stop reading and go install ScreenPipe.

The license question#

Verdict: ScreenPipe is source-available, not open source.

This matters. ScreenPipe moved from MIT to the Screenpipe Commercial License on 2026-06-09. The source is still visible, and you can use it free for personal, non-commercial, educational, or research purposes, plus a 7-day commercial evaluation. Any commercial use of the source requires a paid license. Older MIT-licensed versions remain MIT.

The prebuilt desktop product has three plans: Free, Basic, and Business. Those product plans are separate from the source-code license described above.

Everpaper is proprietary and in private beta. Neither tool is open source today, but ScreenPipe's source-available model means you can read, audit, and build on the code if your use is non-commercial. Everpaper's code is closed. If auditability matters to you, ScreenPipe wins this round.

Pricing#

Verdict: ScreenPipe has a free path; Everpaper has no public price yet.

ScreenPipe's Free plan is $0 with no credit card required and includes a 7-day Business trial, local capture, AI search, meeting notes, 50+ integrations, 10 AI credits/month on Auto only, and limited user-installed scheduled tasks and history; frontier models are excluded. Basic is $25/month or $250/year, with 150 AI credits/month, efficient models (Auto, GPT-5.6 Luna, GPT-5.4 mini), and unlimited user-installed scheduled tasks and history; it still excludes frontier models, though you can use your own key for more. Business is $50/seat/month or $500/seat/year, with 400 AI credits/month, frontier Claude and GPT models, cloud sync, automations on efficient models, unlimited provider usage through your own key, and a managed team workspace.

Everpaper is in private beta with limited spots. No public price is available.

If cost is the deciding factor, ScreenPipe has a documented $0 plan. Its paid tiers are numerically clear: $250/year for Basic or $500/seat/year for Business. Everpaper's price is not public during beta.

Use cases#

ScreenPipe is better for:

• Developers building custom agents on their own capture history.
• Cross-platform teams (Windows, Linux, Intel Mac).
• Researchers who need auditable source for non-commercial work.
• Tinkerers who want to wire Whisper, Ollama, and SQLite into something new.

Everpaper is better for:

• People who want a finished product, not a framework.
• Meeting-heavy workflows that need speaker labels and action items without config.
• Apple Silicon users who want an integrated Parakeet, Nemotron and MOSS workflow.
• Anyone who values ⌥Space Concordance and end-of-day review over API endpoints.

Honest pros and cons#

ScreenPipe#

Pros

• Genuinely local-first with offline mode, PII redaction, and auditable source.
• Platform-agnostic meeting capture: no bot, works with any call software.
• Developer-friendly: SQLite, HTTP API, MCP server, pipes framework.
• Cross-platform: macOS, Windows, Linux.
• Free CLI is capable and useful.

Cons

• License changed from MIT to source-available in June 2026; commercial use requires payment.
• Business features cost $50/seat/month or $500/seat/year.
• Resource usage can be heavy; official docs suggest lowering FPS and using whisper-tiny on older machines.
• The official FAQ reviewed did not document speaker identification or diarization.
• Linux desktop app is partial; Wayland support varies.

Everpaper#

Pros

• Finished product: Day Journal, Meetings and Memory work together out of the box.
• On-device transcription with speaker labels (Parakeet and Nemotron), plus MOSS for Detailed meetings.
• ⌥Space Concordance and Ask the Book make recall immediate.
• Aggressive redaction at capture and before every AI call.
• No account, no API key, no terminal required.

Cons

• macOS 15+ and Apple Silicon only; no Intel, Windows, Linux, mobile, or web.
• Private beta with limited spots.
• Memory readiness gate: 5 days, 300 frames, 60 things, 3 threads before it shows Ready.
• AI features (Ask the Book, Reflection, meeting Ask) wait for a provider.
• No public beta price.

FAQ#

Is ScreenPipe open source?#

No, not since June 2026. ScreenPipe moved from MIT to the Screenpipe Commercial License on 2026-06-09. The source code is visible and free for personal, non-commercial, educational, and research use, plus a 7-day commercial evaluation. Any commercial use of the source requires a paid license. Older MIT-licensed versions remain under MIT. The correct term today is source-available.

Is ScreenPipe free?#

Yes. Free is $0 with no credit card required and includes a 7-day Business trial, local capture, AI search, meeting notes, 50+ integrations, and 10 AI credits/month. Basic is $25/month or $250/year. Business is $50/seat/month or $500/seat/year.

Does ScreenPipe work offline?#

Yes, for capture, OCR, local Whisper transcription, and search. Cloud features (Deepgram transcription, Claude/ChatGPT/Gemini chat, archive, team sync) require network. Everpaper also works fully offline for capture, transcription, and search; AI features wait for a provider.

Which is better for meetings?#

Everpaper detects Zoom, Meet, Teams, Webex, Around, and Whereby links, starts recording with ⇧⌘M, and produces a speaker-labeled transcript. With an AI provider set up, a finished meeting's action items show up in Today's focus; you can also run More → Extract To-Dos into Today on any meeting. ScreenPipe records the machine's screen and audio, but the official FAQ reviewed did not document speaker identification or diarization.

Can I use both?#

This review did not verify simultaneous capture. Both apps can record screen or audio, so test permissions, battery use, and disk growth before running them together.

Final recommendation#

ScreenPipe is the better choice of these two if you want to build. Its local API, pipes framework, and cross-platform support make it a flexible local-first capture tool. The June 2026 license change means it's no longer MIT, but for personal use and research, the source is still there.

Everpaper is the better choice if you want to use. It's a finished product with opinions about how a daybook should work: end-of-day review, Calendar Auto-Start, ⌥Space Concordance, and a Memory that learns over time. The trade-off is platform exclusivity and beta availability.

If you're a developer who wants to experiment on your own data, install ScreenPipe's CLI today. If you're on an Apple Silicon Mac and want a private memory that works out of the box, apply for the Everpaper private beta.

---

Everpaper is in private beta with limited spots. Apply for the private beta; if you're selected, you'll hear back within 48 hours. For more on how the CLI works, see the CLI documentation. For alternatives to ScreenPipe, see our ScreenPipe alternatives roundup.