In Claude Code, a project is now a single conversation: you describe the work and Claude divides it into threads (threads) that run in parallel in the cloud, each on its own branch. Anthropic redesigned Claude Projects on September 17, 2026. A coordinator decides what becomes a thread and tracks its progress, and all threads start from the same repositories, instructions, and memory.
Status as of September 21, 2026: the new Projects are in beta for Pro and Max plans and are being rolled out gradually. They arrive first to accounts that have already used Claude Code cloud sessions and that don’t have existing projects in claude.ai chat or Cowork. They’re not yet available on Team or Enterprise plans. If Projects doesn’t appear in the sidebar of claude.ai/code, the rollout hasn’t reached your account yet; there’s a waitlist.
What changed in Claude Projects?
Until now, a Claude project was a folder: a knowledge base plus a set of chats that shared it. That version still exists in claude.ai chat and Cowork, and works the same as before. According to the Help Center, projects on Pro and Max plans will be updated to the new version when the rollout reaches chat and Cowork.
The new version lives in Claude Code and works differently. You don’t open chats within the project: you talk with a single coordinator conversation, and Claude starts the threads. In Anthropic’s words, Claude “defines the scope of the request, delegates the work, coordinates threads in parallel, reviews the results, and assembles the final deliverable”.
Don’t confuse either of these with the claude project command in the CLI. The documentation indicates it manages the local state of Claude Code for a directory and has no relation to Projects.
How does a project work in Claude Code?
A project has four parts:
-
The project conversation. Claude acts as coordinator. It receives what you send (a bug report, a stack trace, a task list), decides what becomes a thread, and answers quick questions itself. It sees what threads report to it, not each step they take.
-
The threads. Each one is an independent Claude Code cloud session, with its own context window. It works on its own branch, opens a pull request when the work requires it, and reports when done. Threads keep running even if you close your laptop.
-
The shared context. Every thread starts with:
- the project’s repositories and files;
- the project’s instructions (up to 16,000 characters);
- the project memory, a set of files indexed by
MEMORY.mdwhere Claude saves decisions and issues you ask it to remember.
Additionally, threads read the
CLAUDE.md, skills, and plugins from each repository. -
The Overview panel. Shows which threads finished, which pull requests are ready, and which thread awaits your response. It has tabs for Library (your files and those generated by threads), Pull requests, and Routines.
One detail matters: threads don’t take anything from Claude Code’s settings on your machine. Everything a thread needs has to be in the repository, in the project, or in its cloud environment.
How do you use the new Claude Projects?
You work in claude.ai/code, in the Code tab of the desktop app, or in Claude’s mobile app. There are two ways to get started:
- From scratch. Select Projects in the sidebar and then New project. Only the Name is required; Goal and Context (GitHub repositories, files, folders, or Google Drive folders) are optional. Click Create project.
- From an existing cloud session. Open the session menu and choose Continue as a project. Claude reads the session and proposes a configuration.
To work with code you need a repository on github.com, not on GitHub Enterprise Server, GitLab, or Bitbucket. Your connected account must also have push permissions, and the Claude GitHub App must be installed in the repository. The token you get with /web-setup isn’t enough for a project’s threads.
Once the project is created, write its instructions in Project settings > Memory > Project instructions. Indicate which branch the threads start from, how they verify their own work, and what they need to ask you about before doing it.
If you already maintain conventions in CLAUDE.md or AGENTS.md, they apply to each thread. In this guide we explain how Claude Code reads AGENTS.md.
Is it free? How much usage does a project consume on Claude Pro and Max
No: the new Projects require a Pro or Max plan, and they consume the same limits as your other Claude Code sessions, faster. The documentation says it plainly: on Pro, expect to reach your limit sooner on days you have a project running. Three things consume usage:
- Running threads. Each one is a complete session, and multiple can run at once. The fixed limit is 200 new threads per day across all your projects.
- The coordinator. It spends tokens reading reports and deciding what to do next.
- Threads monitoring a pull request. An inactive thread reactivates when CI fails or a review comment arrives. Ask it to stop monitoring if you don’t want that consumption.
A thread that reaches your limit waits and resumes when the limit resets. In Project settings > Usage you see token consumption per thread and per model.
What limitations does it have?
- Not available in the terminal CLI. Nor through Amazon Bedrock, Google Cloud, or Microsoft Foundry. A local session cannot be part of a project.
- Cannot be shared. A project belongs to a single user and has no organization-level controls during beta.
- You can lose work without commit. A thread’s sandbox pauses between turns. If it can’t be resumed, the thread continues from a fresh clone. For long tasks, ask Claude to commit and push as it progresses.
- Local execution hasn’t arrived yet. Running threads on your own machine, behind your own network, is announced as “coming very soon”. In the meantime, work that needs a local database or a service behind your VPN remains for a local session. We explain a way to run Claude Code sessions on your own infrastructure in our self-hosted runner guide.