YNSPIRAClaude Projects
Anthropic
Every time you start a new Claude conversation, you start more or less from scratch. You have to explain what you are working on and how you want things done. Projects removes that repetition.
A project is a workspace where you add files and write instructions that Claude follows in every conversation inside the project. Your style guide, your report template, your tone of voice. You explain it once, and Claude knows it from the first message of every new conversation.
With Projects, Claude behaves more like an assistant who actually knows what you need help with, without you explaining everything again.
Chat, Projects or Cowork?
Regular chat is for one-off questions. You don't need any prior context, and you probably won't return to the topic.
Projects is for work you repeat or come back to. Do you write a newsletter every week? Run the same analysis every month? Have a client where Claude needs to know the background? Projects saves you from explaining the same thing over again.
Cowork is for tasks Claude should handle more independently. Working through folders of files on your machine and producing documents, presentations or spreadsheets. Cowork can be organized into projects too.
How much can you upload?
You can upload roughly 500 pages of text into a project. That is more than enough for most people.
On paid plans (Pro and above) Claude expands the capacity automatically. Once the knowledge base grows past what Claude can hold in its head at once, it starts searching through your files instead and pulls out what is relevant to the question you asked (RAG). You barely notice the difference, and the capacity can grow up to ten times larger.
On the free plan you get five projects, but without that automatic expansion. You are limited to what fits inside the 500 pages.
Sharing, and an important difference from ChatGPT
On Team and Enterprise plans you can share projects with colleagues. They get access to the files and instructions, but the conversations stay private. Each person has their own dialogue with Claude.
This works differently from ChatGPT Projects, where conversations are shared with everyone who has access. In Claude the project is a shared starting point, but what you talk to Claude about is yours alone.
Memory over time
Claude gradually builds up memory per project. It notices how you work, your preferences and the decisions you make. Each project has its own memory, kept separate from other projects. You can view, edit and delete what Claude remembers in the project.
Limitations
New conversations in a project start without the full history of the previous conversation. Claude can draw on the memory it has built up in the project, though, and search through earlier conversations to find relevant context. Uploading files from Google Drive works, but it is a bit clunky.
How to use it
Create a project for each area of work, for example one for marketing and one for product development. Upload the relevant documents (style guides, templates, reports, code) and write project instructions that define the tone, format and rules Claude should follow.
Every conversation you start inside the project automatically has access to the knowledge base and the instructions.
Relevant guides and articles
- Where should your instructions go in Claude?Guide · 7 minRead more →
- How to build an "About me" skill that makes AI much more accurateGuide · 7 minRead more →
- How to write better prompts for Claude Opus 4.8Article · 7 minRead more →
- Skills: when you need them and how to build your first one in ClaudeGuide · 12 minRead more →
- How to set up Claude Projects properlyArticle · 8 minRead more →
- How to get 10x more out of Claude: From chat to working partnerArticle · 10 minRead more →
Features
Knowledge base
Upload documents, code and files that Claude uses as context in every conversation inside the project.
Project instructions
Separate system instructions per project that steer Claude's tone, format and behaviour.
Team sharing
Share projects with colleagues. Three levels: private (not shared), invited people only, or the whole organization.
Retrieval-Augmented Generation (RAG)
When the context grows past 200K tokens, Claude automatically pulls the most relevant parts from the knowledge base using semantic search (RAG).
Supports many file formats
PDF, DOCX, CSV, TXT, HTML, code and images can be uploaded and analyzed.
Does your team need training in Claude?
We run tailored courses and workshops in Claude, adapted to your business and your working day.
Tips for good use
- Use the project instructions actively. Things like always write in Norwegian, use this report template, and refer back to the source material give far better results than repeating it in every conversation.
- Upload style guides, templates and reference documents into the knowledge base. Claude follows them consistently, like a new colleague who actually reads the onboarding material.
- One project per area of work is the rule of thumb. Don't mix marketing and technical documentation in the same project. The context gets muddled.
- Write the instructions as if you were onboarding a new colleague. Think of it as a welcome document. What does this person need to know to do the job? Which tone? What format for the deliverables? Which mistakes should they avoid? The more specific you are, the less you have to correct along the way.
- Update the instructions whenever you notice something isn't working. The first version is rarely right. If Claude repeats the same mistake, wrong tone, forgets the style, slips into English words, write it into the instructions as a rule. The project gets better with every round.
- Check and edit the project memory regularly. Claude builds up memory over time. If it has picked up something wrong or is holding on to outdated assumptions, you can edit or delete it directly. That can be the difference between a project that keeps improving and one that drifts.
Available on
Links and pricing
Included from the Pro plan ($20/mo)
AI is an assistant, not an authority. Always verify what matters.