##### How can you leverage Cursor's existing architecture to build permanent knowledge bases to increase breadth for developers? Cursor is an AI-first code editor that helps build software faster. It's an editor designed for pair programming with AI. It performs well for so many tasks, and has already improved my workflow. I found myself using the web search tool a lot across different sessions and workspaces to import external information or documentation. The issue is that its web search and indexing function is tied to individual chats or workspaces, and the results of the agentic process of searching, scraping, and indexing the webpages to build a knowledgebase is lost between sessions. Agents and Multi-Agent Systems, paired with appropriate web and file tools, can enhance UX and improve cursor’s breadth of capacity. The @Web search is great for small and simple tasks, but why not curate and store the more important knowledge and simply import it into workspaces for recurring tasks? <iframe width="560" height="315" src="https://www.youtube.com/embed/wuSoM0DlXXk?si=dE6BlRcU-umBGCl_" title="YouTube video player" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share" allowfullscreen></iframe> ### Insight Cursor already has a super smooth and effective indexing, vectorization and search capabilities making it possible to manage whole directories of different files with LLMs. This allows Cursor to query and examine relevant code across multiple files. ![[Frame 819.png]] Context management is a major challenge in pair-programming with AI. I think there's a huge opportunity here to build in-editor knowledge management systems. ### Agentic Knowledge Management A multi-agent system with web search and file save capabilities would allow users to enter a query on which developer docs to search and study, and what specifically to focus on. Agents then can search the web for those docs, and scrape a few of the websites to build an understanding and save it as a local file, which can then be drawn in to a workspace and re-indexed to update the workspace with the new information in the knowledge file. ![[Frame 126.png]] When you ask a question across the codebase, Cursor can then see the file in its folder, and choose to read it based on its relevance. ### @knowledgebuilder ![[Frame 870.png]] In the example video, I use this to demonstrate how to leverage Cursor to build or at least start a Crew file. Since this is recurring work I do, having this saved to my permanent files allows me to drag it into Cursor when necessary and add to the knowledge base’s complexity. This enables users to bring in whatever existing research context they want whenever they need to. ### @knowledge ![[Frame 871.png]] ### Expansion Programming, simply put, is really just a means to an end; people want to build! (software). Multi-agent-systems enable recursive agentic tasks to be carried out right in the editor. Over time, the knowledgebase becomes a curated asset that each Cursor user can curate and build to match what they need to use frequently. Game devs might cultivate a deep knowledgebase on game mechanics and add to it with any new updates, where other devs may be working on a smaller project that integrates with a few less common APIs. With this setup, Cursor can just search the local knowledgebase. ![[Frame 820.png]] These can be updated and refreshed whenever, effectively forming a small agentic library system, where agents search, acquire, collect, document, save, store, as well as cull and remove outdated or incorrect information. ![[Frame 821b.png]] The overall idea is that these systems could be operating as a feature enabling a knowledge management system to make Cursor the smartest and best integrated code-editor that's always informed with high-quality and up to date contextual information. This feature makes knowledge expandable and more permanent rather than ephemeral, and enables users to curate a knowledge base that is best suited to their needs. ![[cursor_crew_graph_lg.PNG]] ![[network_lg.PNG]] _____ [[Open File]] [[File Actions]] [[Delete File]] [[Edit File]] [[Indexing]] [[Markdown Reports]] [[Save File]] [[Knowledge as Files]] [[Cursor-Knowledge Folder]] [[Actions]] [[Auto-ReIndexing]] [[Developer Documentation]] [[Information Retrieval Systems]] [[Flexibility]] [[Version Control Integration]] [[Multi-Agent Systems]] [[Pruning, Culling, Removing]] [[Librarian]] [[Collaborative Editing]] [[Curation of Knowledge]] [[Cursor]] [[Learning]] [[Programming Abstraction]] [[Contextual Search]] [[Knowledge Expansion]] [[Web Search]] [[Knowledgebase Creation]] [[Agentic Library Management]] [[Langchain]] [[Automated Summarization]] [[Personalization of Knowledge]] [[Reorganizing and Translating]] [[CrewAI]] [[Semantic Analysis]] [[Knowledge Persistence]] [[Data Ownership]] [[Data Permanence]] [[Persistent Understanding]]