Skip to main content
PromptQuorumBuilt for humans. Structured for AI.
Home/Power Local LLM
Productivity & Knowledge Tools

Screenpipe Review: Local Screen and Audio Memory for You and Your AI Agents

·7 min readBy · Founder of PromptQuorum · Discovery engine for open-weight & open-source AI

Screenpipe is a source-available desktop app for macOS, Windows and Linux that keeps a searchable local record of your screen and audio and exposes it through a timeline, a local REST API and an MCP server. It has a free plan with limited history and paid plans from $21/month. Recording everything you see and hear raises consent and security questions you should settle before installing.

Key Takeaways

  • Runs on macOS 13+, Windows 10+ and Linux; the install page offers a normal download for each.
  • Free plan at $0 for one device with limited searchable history; paid plans start at $21/month (Basic).
  • The code is source-available under a commercial license. Personal non-commercial use is permitted; commercial use needs a license.
  • The maker estimates about 30 GB of disk per month, with configurable retention.
  • PromptQuorum has not tested the app. Hacker News commenters and the founder's Launch HN post raise consent, local security and compliance questions.

📍 In One Sentence

Screenpipe records your screen and audio on your computer and makes that history searchable by you and by AI agents through a timeline, a REST API and an MCP server.

💬 In Plain Terms

Think of it as a searchable memory of everything you saw and heard at your computer. It stores that memory locally by default, but turning on cloud AI, sync or integrations can send data off the device.

📌Note: Recording screen and audio all day can capture other people's voices, passwords and confidential material. Check consent, workplace and legal rules before you install it. This is general information, not legal advice.

Screenpipe is a desktop app that records what appears on your screen and what you say and hear. It stores the history on your computer by default and lets you or an AI agent search it. This review is based on the maker's pages, the GitHub repository, the pricing and install pages and published coverage, checked on 16 November 2026. PromptQuorum has not tested the app hands-on.

Screenpipe Review: Local Screen and Audio Memory for You and Your AI Agents

Overview

Screenpipe is a desktop recorder that turns your screen and audio into a searchable local history.

The app is made by Negentropy Labs, Inc. (doing business as Screenpipe). The README places the company in San Francisco, CA, and the maker's homepage and README state Y Combinator backing (S26). Version 2.7.145, released on 6 November 2026, is the latest GitHub release. The repository had a commit on 15 November 2026, so development is ongoing and no stale-project warning applies.

Per the README, capture uses the accessibility tree first, with OCR as a fallback. It also records audio transcription, speaker information, keyboard inputs and app switches. The result is stored in a local database. You can browse it in a timeline view to scroll back to a past screen, document or conversation.

The feature that separates it from a plain screen recorder is agent access. Screenpipe provides a local REST API and an MCP server, so AI agents and assistants that support these interfaces can query your history. The README shows adding it to an MCP-capable assistant with an npx command. For companies, the maker also describes workflow discovery and an Enterprise workspace.

These descriptions are maker statements. PromptQuorum has not run the app and has not benchmarked its search quality, resource use or accuracy.

How to Get It

You install Screenpipe with a normal per-OS download from the maker's install page, and a terminal route also exists.

The install page offers builds for Mac (Apple Silicon and Intel), Windows, Windows ARM and Linux. The README names a .dmg, a .exe and an AppImage. GitHub release assets were not listed, so the file formats come from the README and the install page. On macOS the app asks for screen and system audio recording, microphone and accessibility permissions; input monitoring is optional.

Platform
How to get it
macOSDownload .zip or .dmg, drag to Applications
WindowsDownload the installer (.exe per README)
LinuxDesktop app (AppImage per README) or source build
CLIRun npx screenpipe record in a terminal

Sign-in prompts appear on the install page, but the sources do not say whether an account is required to use the free plan.

Requirements

Screenpipe's stated RAM and CPU use depends on your settings, and its disk use grows over time.

The maker's figures differ between its site and the README, so treat them as ranges that depend on your settings. No GPU requirement is stated; the changelog only mentions GPU transcription on Windows. On Linux, capture behavior depends on the distribution, desktop environment and permissions, and the sources give no distribution list.

Item
Stated figure
RAM in useSite: 600 MB to 4 GB; README: about 0.5-3 GB
System RAM8 GB recommended (README)
DiskAbout 30 GB/month; retention configurable (site)
CPURoughly 5-20% (README); varies with hardware
Operating systemmacOS 13+, Windows 10+, Linux (x86_64)
GPUNo requirement stated
NetworkLocal capture and search work offline after setup

Cloud AI, sync and connected services need network access.

Features

Screenpipe's core is continuous capture plus several ways to search the result. All items below are maker claims from the website, README, About page and pricing page.

  • Screen and audio recording on the device, building a searchable history.
  • Capture of the accessibility tree, with OCR fallback, plus transcription, speakers, keyboard inputs and app switches.
  • A local database, REST API and MCP server so AI agents and assistants can query the history.
  • A timeline view to scroll back to a past screen, document or conversation.
  • Controls to exclude apps, windows and URLs, and to redact sensitive fields on the device.
  • Optional encryption at rest.
  • Workflow discovery and editable process documentation from recorded work. The maker says workflows the tool produces need review, and the pricing page says automatic workflow discovery requires Enterprise.
  • An Enterprise workspace with MDM, SSO and audit logs.
  • Questions about your work are answered by a model that depends on the plan: Auto (Free), efficient (Basic), frontier (Business) and custom (Enterprise), per the pricing page.

Privacy and Data Flow

Raw capture history is stored locally by default, but several features can send context off your device.

The maker's AI memory privacy checklist says cloud AI, transcription, sync, integrations and managed team storage are separate data paths. It tells users to check their selected transcription and AI settings rather than assume defaults. The changelog (v2.7.79) says the ai-gateway uses GPT-6 Luna for auto defaults. Review which AI path your plan uses before relying on local-only behavior.

In the Launch HN post, the founder says the most useful criticism concerned recording consent, local security, CPU usage and signal-to-noise. Commenters on that thread worried that a compromised agent could query the full history through the local API on port 3030. They also noted that even local-only storage can start a compliance discussion for teams. A commenter in a 2024 HN thread said the tool opens a huge can of worms for privacy.

One reviewer on Medium says the PII scrubbing is good but not bulletproof, and that users must handle backups and disk encryption themselves. A blog post by LUCI, a competing product, reports that the incognito-window skip only checks the focused window (GitHub issue #6198). PromptQuorum did not check the current status of that issue.

Account requirements, telemetry and a privacy policy URL are unverified; see the section below.

Who Should Use It

Screenpipe fits people who can lawfully record their own computer use and want it searchable.

  • People who want a searchable record of what they saw and heard on their computer, with local storage by default.
  • Developers who want AI agents and coding assistants to query their work history via MCP or a local API.
  • Teams with recurring cross-app work who want to map workflows and are prepared to agree capture and retention rules first.

What We Could Not Verify

Screenpipe is a poor fit in some situations, and several details remain unverified.

  • Not a fit: anyone who cannot or should not record screen and audio, for example because of consent or compliance rules or sensitive work.
  • Not a fit: users who need an OSI-approved open-source license or free commercial use of the code.
  • Not a fit: machines with small disks or low specs that cannot allow for roughly 30 GB per month of storage and 0.6-4 GB of RAM.
  • Not a fit: teams that need shared context across teammates below the Enterprise tier. Shared team context, automatic workflow discovery and cloud API access require Enterprise.
  • Unverified: whether an account is required. The install page mentions sign-in prompts, but no source states a requirement.
  • Unverified: whether the app sends telemetry or analytics. No source states this.
  • Unverified: a privacy policy URL. The maker's checklist refers to a privacy policy and security overview but gives no link, so the privacy link in this review is the checklist post, not a policy.
  • Unverified: any GPU requirement, and the exact limits of the free plan. The pricing page says only "limited" history and scheduled workflows, without numbers.
  • Unverified: the license history. A third-party review says releases before the maker's June 2026 license change remain MIT, and a directory listing elsewhere says MIT. Only the Screenpipe Commercial License file in the repository was used here.
  • Unverified: the maker's company history. An older README names Mediar, Inc., and a third party says it was formerly Mediar AI. No primary source confirms a rename, and the sources differ on the founding year.
  • Unverified: customer logos on the homepage are the maker's own statement. The Launch HN post was written by the founder, so only its comment thread counts as third-party.
  • Unverified: the current status of the incognito-window issue reported by LUCI's blog.
  • Unverified: the headquarters location comes from the README and a Y Combinator profile only; no imprint page was read. No app store listings were found, so there are no store ratings.
  • Unverified: an old repository mentions a $400 lifetime purchase, and one review says $25/month. This review uses the maker's current pricing page.

Screenpipe vs Alternatives

Screenpipe records screen and audio and exposes an MCP server and a REST API. PromptQuorum has not compared it with the alternatives hands-on. The alternatives below come from third-party descriptions and comparison sites.

Screenpipe

Noted in sources:
Source-available; screen and audio; Mac, Windows, Linux
Also mentioned in Screenpipe (1)

Also mentioned in:

Microsoft Recall

Noted in sources:
Windows 11 Copilot+ PCs; no audio (third party)

OpenRecall

Noted in sources:
Open-source (AGPLv3) local recall tool

Windrecorder

Noted in sources:
Open-source screen recall tool for Windows

LUCI

Noted in sources:
Proprietary consumer recall app (third-party comparison sites)

Compare in directory

FAQ

Is Screenpipe free?

Partly. There is a free plan at $0 for one device with limited searchable history and limited scheduled workflows. Basic is $21/month or $250/year, Business is $42/seat/month or $500/seat/year, and Enterprise is custom-priced. Students, researchers and faculty get 50% off the first month of an individual subscription, per the pricing page.

Do my recordings leave my computer?

Raw capture history is stored locally by default. Cloud AI, transcription, sync, integrations and managed team storage are separate data paths that can send context off the device, depending on your settings. The maker's own checklist advises checking those settings rather than assuming defaults.

Which operating systems does it support?

macOS 13+, Windows 10+ and Linux (x86_64). The download page offers builds for Mac (Apple Silicon and Intel), Windows, Windows ARM and Linux. Linux capture behavior depends on distribution, desktop environment and permissions.

Is Screenpipe open source, and can I use it commercially?

The code is source-available under the Screenpipe Commercial License, which is not an OSI open-source license. Personal non-commercial use is permitted, and commercial use requires a license. Commercial source builds require Enterprise, per the pricing page.

Can other AI agents and assistants use my Screenpipe history?

Yes, per the maker. It provides a local REST API and an MCP server, and the README shows adding it to an MCP-capable assistant with an npx command. Some Hacker News commenters worry that a compromised agent could query the full history through the local API, so consider which agents you connect.

Verdict

Screenpipe is worth a look if you want a searchable local memory of your computer work and can settle the consent and security questions first.

On paper it covers a lot: local storage by default, a timeline, a REST API, an MCP server, exclusion and redaction controls, optional encryption at rest, and a free plan to start on one device. Development is ongoing, with a release on 6 November 2026 and a commit on 15 November 2026.

The trade-offs are real. The license is a commercial source-available one rather than open source, and history length depends on the plan. Disk use is about 30 GB per month by the maker's estimate. The cloud features mean it is not fully local in every configuration. Account requirements and telemetry are unverified. PromptQuorum has not tested the app. Try it on a non-sensitive machine, set retention and exclusion rules, and review the AI and sync settings before you rely on it.

Sources

  • Screenpipe website — maker's description of recording, timeline, REST API, MCP server, exclusion controls and Enterprise workspace, checked 16 November 2026.
  • Screenpipe GitHub repository — repository, last commit and project description, checked 16 November 2026.
  • Screenpipe README — capture method, install formats, resource figures, CLI and MCP setup command, checked 16 November 2026.
  • Screenpipe README (main branch) — San Francisco, CA location, checked 16 November 2026.
  • LICENSE.md — Screenpipe Commercial License title, checked 16 November 2026.
  • Release app-v2.7.145 — latest release version and date, checked 16 November 2026.
  • Install page — per-OS downloads and permissions, checked 16 November 2026.
  • Pricing page — plans, prices, plan limits, storage and RAM estimates, checked 16 November 2026.
  • AI memory privacy checklist — maker's description of local and cloud data paths, checked 16 November 2026.
  • Changelog — ai-gateway note in v2.7.79 and Windows GPU transcription mention, checked 16 November 2026.
  • About page — workflow discovery description, checked 16 November 2026.
  • aicoolies review — third-party license commentary, checked 16 November 2026.
  • Medium technical teardown — PII scrubbing, backups and disk encryption comments, checked 16 November 2026.
  • Launch HN: Screenpipe — founder post and commenter concerns about consent, the local API and compliance, checked 16 November 2026.
  • Earlier Hacker News thread — 2024 commenter privacy concern, checked 16 November 2026.
  • LUCI blog post — report on the incognito-window issue from a competing product's blog, checked 16 November 2026.

← Back to Power Local LLM

MCP