the problem a project solves

A plain Claude.ai chat starts from zero every time — no memory of the client brief you pasted in last week's conversation, no awareness of the style guide you explained twice already. A Project is a persistent container: files and instructions you set up once, and every new chat inside that project starts already knowing them.

project knowledge

Upload the documents that define the recurring context — a style guide, a product spec, past examples of the output you want, reference data. Every conversation inside the project can draw on all of it without you re-attaching anything or re-explaining background.

custom instructions

Alongside files, a project has its own standing instructions — the equivalent of a system prompt that applies to every conversation inside it: tone, format, constraints, things to always do or never do. This is the same idea as a CLAUDE.md file for Claude Code, just for the chat-app side of Claude instead of a codebase.

conversations stay organized, not merged

A project doesn't create one giant shared conversation — each chat inside it is still its own thread, just with the project's files and instructions available from the first message. That matters for a subtle reason: two conversations in the same project don't see each other's history, only the shared project knowledge. Use one long-running chat per sub-topic (one for drafts, one for feedback rounds) rather than one enormous thread that never ends.

availability and limits

Projects are available on every plan, including free — free accounts are capped at a handful of projects, and larger project knowledge (beyond what fits directly in context) requires a paid plan to actually search across everything uploaded rather than just what fits in the window. Team and Enterprise plans add sharing and permissions on top, so a project can belong to a whole team rather than just one person.

when a project is worth setting up

SituationUse
A one-off question or taska plain chat — a project adds setup cost with no payoff
Recurring work with shared context (a client, a course, a long-running piece of writing)a project
Actually reading/editing files in a real codebaseClaude Code, not a project — projects don't execute code or touch your filesystem

related topics

Claude Cheat Sheet — where projects fit relative to Claude Code and the API.
Effective Prompting with Claude — writing the custom instructions that make a project actually useful.
Claude's Different Kinds of Memory — how project knowledge and custom instructions compare to CLAUDE.md, auto memory, and the rest.

reference

support.claude.com — what are projects