Back to Updates

Agentic Long-Term Memory and Meeting Prep

Product

The agentic era begins — multi-turn agentic Long-Term Memory, Meeting Prep, Reflection Mode, and Google Calendar.

6.0.0 changes the architecture. Agentic Long-Term Memory Engine replaces the single-shot Q&A engine with a multi-turn reasoning agent that follows threads, cross-references context, and builds toward complete answers. That foundation powers the first summary that takes action — Meeting Prep reads your calendar, assembles structured pre-reads, and can schedule prep time for you.

What you can do now that you couldn't before

Have real conversations with your memory

Agentic Long-Term Memory enables multi-turn reasoning across your summaries, events, people, and beyond. The agent follows threads, cross-references context, and builds toward complete answers instead of guessing in one shot. A full toolbox — memory search, web search, calendar, filesystem, and browser history — lets the agent actively gather evidence during any chat or summary.

Walk into every meeting prepared

Meeting Prep connects to your Google Calendar, looks ahead at upcoming meetings, cross-references each with your Long-Term Memories, and generates structured pre-reads. With your permission, it can schedule prep time on your calendar so the work to get ready is actually on your schedule.

Let the agent reflect and self-correct

Reflection Mode gives the agent metacognition — the ability to step back, evaluate its reasoning, and adjust course in real time. The result is better answers even from leaner models, and deeper insights when you ask Pieces to reflect on weeks, months, or quarters of work.

Edit summaries with AI, right inline

Select any text in a Workstream Summary, choose edit with AI, and review suggested changes in a clean diff view. Rework an entire summary with a single prompt — make it shorter for Slack, or rewrite it for an exec audience — all grounded in the original memories.

See the true cost of agentic work

Per-conversation token usage tracks input, output, reasoning, and cache, including tool calls. BYOK and enterprise deployments get clean visibility into provider usage.

Manage your LTM data precisely

Clear data by time period, capture modality — vision, clipboard, audio — or specific application, or combine all three. You can also block apps from capture before Pieces ever encounters them.

Set up MCP in one click

Configure the Pieces MCP Server for Cursor, VS Code, Claude Desktop, and more directly from settings — no JSON editing required.

Who benefits most

  • Everyone: agentic chats and summaries deliver fundamentally better answers.
  • PMs, managers, sales, and consultants: Meeting Prep turns a full calendar into structured pre-reads.
  • Anyone sharing summaries with teammates, clients, or stakeholders: edit with AI before you send.
  • Privacy-conscious users and regulated industries: granular LTM data management and per-app capture controls.
  • Developers using Cursor, VS Code, and Claude Desktop: one-click MCP setup.