Screenpipe is an open-source screen and microphone recording library designed to turn everyday computer work into searchable AI memory and to enable any AI model or agent to make use of this historical data.
Purpose
The core goal of Screenpipe is to continuously record what users do on their computers every day (screen images, audio, application activity), forming a local, searchable history, and then letting AI use this history to understand workflows, generate summaries, follow-up items, and automation suggestions. It is officially described as "summarizing your daily work into AI memory" and can "discover repetitive workflows, record steps, and find opportunities for cross-application automation."
Main capabilities
- Records the screen and microphone around the clock, capturing screen, audio, and application history.
- Combines screen frames with OCR fallback, accessibility data, and audio transcription as supplementary context beyond text events.
- Supports searching what you have seen or heard, and looking back at a specific screen, document, or conversation from the day.
- Allows AI agents to query local work history to understand work and take action, for example by comparing the results when asking the same model with "work history" versus "no work history."
- Can surface repetitive cross-application work patterns, such as invoice emails, amount reconciliation, approval requests, and updating tracking sheets, and provide reviewable automation suggestions.
- Provides a local database, REST API, MCP server, and reviewable source code, so historical memory is not tied to a single model.
- Supports excluding specific applications, windows, and URLs, and can redact sensitive fields on-device.
- Provides an enterprise workspace for centrally managing devices, identities, policies, storage, and audit records.
How to use it
The official page shows support for macOS, Windows, and Linux, and it is source-available. After installation, it is recommended to first clarify the scope to capture and exclude; existing tools can be connected once and then made available to approved workflows. Historical data is stored locally by default and can be called by Claude, Codex, or custom agents through the local database, REST API, or MCP server. The Q&A on the page covers capture, privacy, reliability, CPU/battery/storage usage, supported platforms, cost, and enterprise deployment, but the specific answers are not elaborated in the provided materials.
Use cases
- Use meetings and daily work to draft summaries and follow-up items.
- Discover repetitive cross-application processes and evaluate automation opportunities, such as invoice reconciliation and payment tracking.
- Look back at previously viewedscreens, documents, or conversation content, and continue work from where it was interrupted.
- Let AI agents obtain context and perform actions based on the work history of an individual or team.
- Enable enterprises to centrally deploy and manage policies, storage, and auditing while keeping the original history localized.
Limitations
The materials clearly state that the original history is stored locally by default, but information about cost, specific pricing, CPU/battery/storage consumption, and other unelaborated content in the Q&A is not yet provided. The page also mentions that sensitive work can be avoided from being recorded by excluding applications, windows, and URLs and by redacting sensitive fields on-device, but the specific configuration methods are not explained in the materials.
