Pieces

Pieces builds an AI memory layer that captures desktop work context so developers and other knowledge workers can retrieve past information and prepare their next task. Pieces Desktop organizes research, chats, notes, and meetings into a searchable timeline, with memories stored on-device by default. Users can generate summaries or ask questions across their activity. Agentic Long-Term Memory follows threads through multiple retrieval and reasoning steps; Meeting Prep connects Google Calendar events with stored context to create pre-reads. The Pieces MCP Server makes that memory available to assistants such as Claude, Cursor, and Codex.

Founded in 2020 by Tsavo Knott, Mack Myers, and Mark Widman, Pieces initially targeted developers; Knott serves as co-founder and CEO. Its engineering approach uses small, specialized nano models running locally to extract metadata, rank content, and surface relevant memories. The company fine-tunes open-source models for these focused tasks, reducing reliance on cloud calls for routine context processing. Users can pause capture, exclude applications or websites, and selectively delete stored activity.

In July 2025, the company reported more than 100,000 active users running Pieces locally. Flat Capital invested approximately SEK 10 million in August 2025 as part of a capital raise. Its enterprise offering adds organization-wide capture policies and support for customers’ own API keys.