Using Claude Projects
Over the holidays I've been experimenting with using Claude projects for personal knowledge management in order to better integrate Claude into my workflows both personal and work wise.
What are Projects? A key benefit of projects is the ability to provide > context for your chats with Claude. You can upload relevant documents, text, code, or other files to a project’s knowledge base, which Claude will use > to better understand the context and background for your individual chats within that project. > I found that personally the largest ux obstacles to me > frequently consulting an LLM is the context limit and the inconvenience of uploading all the relevant data into a single chat instance. > Enter Claude > projects which allows you to setup discrete workspaces with custom instructions and a shared knowledge based of files. > Anthropic.com
That's great, because now I can set up specific topics and upload the tiny subset of relevant context that I might want to reuse frequently.
That still leaves the upstream problem, which is that one needs to have the relevant context available as easily parsable, shareable text files. In a true human conversation, the relevant context could be spread across disparate blog posts, tweets, books and forums.
So in the spirit of Text is the Universal Interface, I began obsessively compiling raw .txt and .md of any interesting podcasts and podcasts.
The JinaAI api was super helpful, but it was just a few too many clicks for me to copy the url, prefix it with jina's endpoint, copy the output from the browser to a text file. So with Claude's help I threw together a tiny node CLI to instantly download a markdown file given a url.
npx getmd <YOUR_TARGET_URL>
The context along with some personal background about myself (my job, my city, existing knowledge, areas of interest, etc, etc) injected via custom instructions elevates the quality of the response.
When asked about a new blogpost, Claude blends the new content with the wider context and salient points that might be most relevant to me:
Another useful trick I found is to "summarise" prior conversations into text files and add those back to the context. That way talking to Claude in the context of a Project feels like a series of related conversations. And since its all text files it can easily be synced/exported/remixed for future use.
