screenpipe: A local-first application that records your computer work and turns it into reviewed workflows you can hand to any agent or automation tool.
What problem it solves
People need to find repeated computer work, document its steps, decisions and exceptions, and decide what should run before handing it to an agent or automation tool. Screenpipe addresses this by recording work locally, turning that history into reviewable workflows, and allowing the automation to run elsewhere.
Key features
Workflow discovery from real runs — Observation is turned into a discovery report that names the trigger, systems and handoffs, repeated steps and decisions, frequency and active time, friction and exceptions, one candidate improvement with a human owner, and the evidence gaps that need another observation.
Source: README
Event-driven screen and audio capture — Capture is triggered by meaningful OS events rather than by a fixed interval, pairing each screenshot with the accessibility tree and falling back to OCR when accessibility data is unavailable. System audio and microphone input are transcribed locally with Whisper or in the cloud with Deepgram, with speaker identification and diarization.
Source: README
Local-first storage and AI search — Frames, audio, transcripts and the search index are stored on the device by default, and local capture and search work offline after setup. History is queryable with natural language across accessibility text, OCR fallback text and audio transcriptions, with keyword full-text search underneath.
Source: README
Pipes as scheduled AI agents — Pipes are scheduled AI agents defined as markdown files, each with a prompt and schedule, and developers can create them in ~/.screenpipe/pipes/. Per-pipe YAML frontmatter controls app, window, content type, time and endpoint access, enforced at skill gating, agent interception and server middleware rather than by prompt.
Source: README
Open interfaces for agents and scripts — The capture history is exposed to AI assistants through an MCP server, a REST API on localhost:3030, and a JavaScript/TypeScript SDK, so an existing agent or automation tool can read screen history, frames and transcriptions.
Source: README, packages
Quick start
Copied verbatim from the project's README.
npx screenpipe recordnpx screenpipe setup
# or
claude mcp add screenpipe -- npx -y screenpipe-mcp@latestGET http://localhost:3030/search?q=meeting+notes&content_type=all&limit=10Who it may suit
Inference: the project may suit operators, consultants and knowledge workers who need to find and document repeated work before automating it, and developers who want to give AI coding assistants context about what they are working on. It may also suit teams deploying AI agents with deterministic per-pipe data permissions.
Topics to discuss
How are the discovery reports validated so that a workflow handed off for automation reflects the real process rather than one session's exceptions?
Which retention, encryption and sync settings should a team default to before rolling local capture out across devices?
What should the data-permission model cover next to keep pipes deterministic without blocking legitimate automation?
Written by ReadmeX AI from the project's README, source and public GitHub data; it may contain mistakes.
Read the original README https://github.com/screenpipe/screenpipe#readme
ReadmeX opened this community for a trending project; it is not run by the project. Maintainers can claim it from the community page.
Comments
…