screenpipe | Long-term memory for teams and AI agents

What

Screenpipe is a local-first desktop capture platform that records screen frames, accessibility data, and audio across macOS, Windows, and Linux. Built for knowledge workers, developers, and teams, it solves context fragmentation by transforming daily computing activity into a searchable, long-term memory layer for AI agents. The software positions itself as an open, privacy-centric alternative to cloud-tethered activity recorders by keeping raw data local and accessible via an open API.

Features

• Multimodal Local Capture: Ingests screen activity via OCR fallback, accessibility events, microphone input, and system audio into a local SQLite database without uploading raw recordings to the cloud.
• Local API and MCP Server: Exposes historical context through standard endpoints and a Model Context Protocol (MCP) server, allowing local and external AI agents to query past user activities.
• On-Device Redaction and Privacy Controls: Filters sensitive windows, applications, and URLs prior to capture, while automatically redacting sensitive text patterns like passwords and payment cards.
• Event-Triggered Automations ("Pipes"): Runs background workflows based on triggers—such as meeting terminations—to summarize conversations, extract action items, and push updates to external tools like Linear, Slack, or Notion.
• Enterprise Workspace Management: Provides centralized administration features, including SSO, MDM compatibility, and audit logging, to facilitate controlled corporate rollouts.

Helpful Tips

• Define Capture Boundaries Prior to Rollout: Configure rule-based exclusions for sensitive internal portals, personal messaging apps, and password managers before initiating continuous capture.
• Benchmark Local Storage and Compute: Given continuous audio transcription and OCR processing, evaluate local device storage utilization and battery consumption across target operating systems before wide deployment.
• Leverage the MCP Interface: Use the local MCP server rather than building bespoke extractors to ground your current AI assistants directly in desktop context.

OpenClaw Skills

The provided documentation contains no native integration with OpenClaw. Plausibly, an OpenClaw agent could connect to screenpipe's local REST API or MCP server to inspect recent desktop and meeting events, enabling a hypothetical workflow where the agent automatically extracts action items from completed calls and creates tracked tasks across project management tools. If realized, this capability could significantly reduce manual administrative reporting and data entry for sales and operations teams.