Pieces is a desktop memory layer that runs in the background and records what you do across applications, then lets you query that history in natural language. Its Long-Term Memory engine (LTM-2.7), running inside the PiecesOS service, forms memories every two seconds from the focused application, from what you copy and paste, and, on an opt-in basis, from meeting and conversation audio; it condenses them into summaries every 20 minutes.
Captured context covers email threads in Gmail and Outlook, browsing in Chrome, Arc, and Safari, Slack conversations, notes, and meetings. The Timeline in the desktop app shows this activity chronologically, with search and filters by time, source, topic, person, or tool, and one-click outputs such as standup updates, morning briefs, meeting prep, and day recaps generated from your own history.
Conversational search answers questions over that memory with a choice of model families, including Claude, Gemini, and ChatGPT, in Fast, Balanced, or Extra Thinking modes. The same memory is reachable from IDE and browser integrations and through an MCP server, so coding agents and other clients can pull your recent context.
Features
- Background capture: memories formed every two seconds from the active app and clipboard, with audio capture as opt-in
- Timeline: chronological view of captured activity with search and filtering
- Scoped prompts: ask about a period, topic, person, or tool and get answers grounded in your history
- Generated recaps: standup updates, morning briefs, meeting prep, and day summaries
- Source coverage: Gmail, Outlook, Chrome, Arc, Safari, Slack, notes, and meetings
- Model choice: Claude, Gemini, and ChatGPT families with Fast, Balanced, and Extra Thinking modes
- MCP server and integrations: expose memory to IDEs, browsers, and agent clients
- PiecesOS: the local service that runs capture and summarisation and serves the app and integrations