I understand that Cursor collects various information in a local vector database.
My hypothesis:
This includes information regarding the current task, but also higher-level data—insights or experiences—that could be useful beyond just that specific task.
Does Cursor have any mechanisms that would allow for the extraction and sharing of this data?
Something akin to “skills,” but not in the form of theory—rather, based on practical experiences or lessons learned.
Example: The DRY principle—not as a dry guideline, but as a vector representation of experiences resulting from failing to adhere to it.
Hey, thanks for the kind words and the great request. There’s a small clarification about the premise.
Codebase index never stored “experience” or insights. It only stored vector embeddings of code chunks for semantic search, and it was server-side, not in a local database. It didn’t accumulate any lessons learned. Also, the codebase indexing settings are now removed from the UI. Agents use Instant Grep instead of a separate semantic index, so a semantic or embeddings vector DB that could “accumulate” anything is no longer being built. There is still a local grep index, but it’s just an index for fast code search, not a storage of “experience”. Staff explanation about this change: What do you think about Cursor removing the codebase indexing settings?
What you’re describing, practical experience packaged for reuse and sharing, is done in Cursor via plain text, not vectors:
Skills .cursor/skills/<name>/SKILL.md are for encoding practical, experience-derived procedures. These files live in the repo, so they’re versioned in git and shared with the team. For wider distribution, there are Plugins. Docs: Agent Skills | Cursor Docs
Rules .cursor/rules, AGENTS.md are for rules and team conventions. Docs: Rules | Cursor Docs
A practical trick that covers almost everything you want: at the end of a session, ask the agent to “distill what you learned in this session into a skill/rule”. The agent will write the SKILL.md and turn real experience, like diagnosing a DRY violation, into a reusable, shareable artifact.
If you want fully automatic extraction of “experience” into a ready-to-share format, post it as an idea in Ideas → Feature Requests and I’ll pass it to the team. Let me know if any of this doesn’t fit your use case.