Pieces

Pieces solves a problem every developer has: 'I saw the answer to this yesterday, I just don't know where.' It runs as a local desktop app (PiecesOS) that continuously and privately captures the context you actually work in — the websites you read, the Stack Overflow snippets you copy, the terminal commands you run, the docs you skim, the meetings you sit through — and turns it into a searchable, queryable long-term memory layer. Plugins for VS Code, JetBrains, Chrome, Obsidian, Microsoft Teams, and the system tray let you ask things like 'what was that error I hit yesterday?' or 'summarize my last hour' and get grounded answers from your own activity, not a generic LLM. All processing can run locally on a model of your choice via Ollama, Llama 3, or another local backend so sensitive code never leaves the machine; you can also point Pieces at cloud models when you want frontier-model quality. Pieces ships an official Model Context Protocol server: any MCP-aware agent — Claude Desktop, Cursor, Cline — can plug into your Pieces memory and use your captured context as a tool. The free tier covers individual use; a Pro tier (~$34.99/month) adds advanced models, larger memory windows, and team features.