[Feature Request] Living Artifacts Panel & In-Session Semantic Search for Heavy Agentic Workflows

Ideas Feature Requests
,
1

Feature request for product/service

Cursor IDE

Describe the request

1. Problem Statement & Pain Points

When running heavy agentic development workflows in Cursor for large enterprise codebases (handling complex SSH, multi-tenant databases, OCR/IoT integrations):

  • Context Decay: In long sessions (24-48 hours), critical decisions (architecture rules, logic fixes) get lost in past chat history.
  • Non-Persistent States: Standard chat UI treats core system architecture decisions the same as disposable Q&A.
  • No In-Session Semantic Search: Hard to locate or jump back to specific code snippets or logic discussed hours ago.

2. Proposed UX/UI Architecture (3 Pillars)

  • Pillar 1: Living Artifacts Panel (Right-hand Column)

    • A dedicated active board alongside the main chat.
    • When key technical agreements are reached, the AI extracts them into pinned “Active Task Cards” that stay visible until explicitly completed or dismissed.

  • Pillar 2: SSOT Auto-Sync (Google Drive / Workspace)

    • Seamless auto-compilation of session notes and technical blueprints directly to linked cloud documentation (Google Docs/Sheets).

  • Pillar 3: In-Session Semantic Search

    • Fast internal search (Ctrl+F equivalent for session context) to instantly highlight and jump to past code blocks, variables, or decisions.

3. Full Detailed Proposal

We have documented the complete architectural blueprint and technical diagrams here:

Feedback and discussion from the Cursor engineering team and community are welcome!

4

Until something like this ships, here’s what has worked for me on long sessions. Keep a decisions.md in the repo and ask the agent to append to it whenever an architecture rule or fix gets settled. Anything that must always apply goes in .cursor/rules so it rides along on every chat. Then instead of letting one chat run 24-48 hours, start a fresh one every few hours with “read decisions.md and continue”. You lose far less to context decay, and grep over that file covers a lot of the in-session search need.