Projects and Memory are the two features that make Claude Desktop fundamentally different from the web app at claude.ai. Projects give conversations persistent shared context. Memory gives Claude a continuously updated record of what it knows about you. Together, they mean you stop re-explaining yourself — to any conversation, in any session.
Projects
A Project is a named container for related conversations. Every conversation inside a project shares the same base context: the files, instructions, and background you attach to it.
Creating a project
- In the Claude Desktop sidebar, click New Project
- Give it a name that reflects the work it covers (e.g. "Client: Henderson", "Research: Leadership", "Content Pipeline")
- Optionally add a project description — this becomes part of every conversation's context
- Attach files: PDFs, Word documents, spreadsheets, text files — anything you want Claude to reference across all conversations in this project
What to put in a project
- Client background documents and briefs
- Style guides or tone-of-voice instructions
- Research papers or reference materials
- Custom instructions specific to this area of work
- Templates Claude should follow for outputs in this project
Project instructions
Each project has an Instructions field (separate from file attachments). Treat this like a CLAUDE.md for that project — the role Claude plays, the conventions to follow, the format of outputs. Instructions apply to every conversation in the project automatically.
Good uses for projects
- Client work — one project per client; attach the brief, background documents, and any templates; all conversations about that client have full context
- Research areas — attach key papers and a research brief; ask questions, synthesise findings, draft writing — all in one context
- Recurring workflows — a "Social Media" project with your brand voice guide and content calendar attached; every post drafted in this project follows the same standards
- Personal assistant — a general project with your preferences, frequently used templates, and background on your work
Memory
Memory is a dynamic, Claude-managed layer that persists facts across conversations. Unlike Projects (which you configure explicitly), Memory updates automatically as Claude learns things about you.
What gets remembered
- Preferences you state ("I prefer bullet points over numbered lists for task summaries")
- Facts about your context ("I work in a team of five; the others are Sarah, James, Priya, and Tom")
- Things you explicitly ask Claude to remember ("remember that my main client is Henderson Construction")
- Corrections you make to Claude's defaults
Viewing and editing memory
Access memory through Claude menu → Memory or the memory icon in the interface. You can see what Claude has stored, edit any entry to correct it, or delete entries that are no longer accurate. Memory is not hidden — you have full visibility and control over it.
Memory is per-account, not per-project
Memory applies across all your conversations in Claude Desktop — it is not scoped to individual projects. If you need context to apply only within a specific project, put it in the Project Instructions field instead. Memory is for facts that are globally true about you; Project Instructions are for context specific to a body of work.
Projects + Memory together
The practical effect of running both: a new conversation in a project opens with project files and instructions already loaded, memory-stored preferences already applied, and Claude already knowing the relevant background. You ask the actual question rather than setting up context first. That's the shift — from context-setting to working.