Guide
How to use Claude Projects to organize your work
A Claude Project is a workspace: its own chats, its own uploaded knowledge, and its own separate memory, so client work never bleeds into product planning. Here's how to split them, what to put in them, and the one habit that makes a project compound instead of sprawl.
Projects are the single most underused feature in claude.ai. Most people run everything in one endless stream of chats, then wonder why Claude keeps forgetting which client they mean. I run a new business build every week, each in its own project, and the structure below is what survived.
What is a Claude Project?
A project bundles three things: the chats you have inside it, the knowledge files you upload to it, and, with memory enabled, its own memory space that stays separate from your other projects. That separation is the point. Claude inside my CalcWolf project knows about calculators and SEO structure; it doesn't drag in my newsletter pricing debate.
How should you split projects?
One project per real-world workstream. For me that's one per weekly build, plus one for SideRoad itself. For a freelancer it's one per client. The test: if two things never need to share context, they're two projects. Resist the "General" mega-project; it recreates the exact soup projects exist to fix.
What belongs in project knowledge?
Two kinds of files. Stable references: brand guidelines, a product spec, the voice doc, things Claude should always have in view. And one living file: the running record of decisions and open tasks for that project. Mine is jotthread.md, maintained by JotThread, so every chat in the project starts already knowing what's open and what got decided. Keep knowledge lean; every file you upload rides along in context, and a bloated project is a slow, distracted one.
The habit that makes it compound
Project knowledge holds what you uploaded, not what changed since. So the loop is: capture during the conversation, get the updated file back, save it into the project. With JotThread that's /jot, a one-line confirmation, and a file card to drop back into knowledge. Do that for a month and the project becomes something rare: an AI workspace with an actual institutional memory you can open in a text editor. For what Claude remembers on its own versus what belongs in your file, see the memory guide.
Common questions
How many projects should I create?
One per real-world workstream: a client, a product, a venture. If two things never share context, they're two projects. If you keep re-uploading the same file to both, they might be one.
What's the difference between project knowledge and memory?
Knowledge is files you deliberately add and can read yourself. Memory is context Claude builds from your conversations, viewable and editable in settings, kept separate per project. Knowledge is your record; memory is Claude's context.
Do project files update themselves?
No. Project knowledge holds the version you uploaded. When a working file changes, save the updated version back, which is exactly the loop JotThread's file card is designed for.
Give every project a memory it can keep.
JotThread files each project's decisions and tasks into one Markdown file that lives in your project knowledge. $29 once.