Claude Code Projects Turn One Conversation Into Parallel Cloud Sessions

Anthropic is rolling out a redesigned Projects experience in Claude Code. The beta gives a project one coordinating conversation that can delegate work to parallel cloud sessions, keep shared memory, and continue working after you close your laptop. (Anthropic's announcement; Claude Code Projects documentation; ClaudeDevs on X)

This is a different shape from a folder that simply holds several chats. You describe the work in one conversation, and Claude decides which requests should stay in that conversation and which should become threads. Each thread is a full Claude Code cloud session with its own context window and branch. (Claude Code Projects documentation)

Key Takeaways

• One coordinator: A long-running project conversation routes tasks and tracks the threads it starts.
• Parallel cloud work: Threads run in parallel as separate Claude Code sessions, each handling one part of the work.
• Shared context: New threads start with the project's repositories, files, instructions, memory, connectors, plugins, and cloud environment.
• Work continues away from the desk: Cloud threads keep going after you close your laptop, and you can check on them from your phone.
• Beta and account-gated: Projects are rolling out gradually to Pro and Max users; Team and Enterprise access is not available yet.
• Higher usage pressure: Every thread is a full session, and the coordinator also consumes tokens, so a project can reach plan limits faster.

One Conversation, Many Threads

The project conversation is the control surface. Claude receives bug reports, task lists, stack traces, updates, and quick questions, then routes each message based on the work involved. A quick question can be answered in the conversation; a new piece of work can be sent to a new thread or an existing one working in the same area. (Claude Code Projects documentation)

Threads are the workers. Each one runs as a separate cloud session on its own branch, reports back when it finishes, and can open a pull request when the work calls for one. Threads can also split delegated work into subagents, loops, and workflows. If two threads change the same code, the result can be a merge conflict just as it would with other pull requests. (Anthropic's announcement; Claude Code Projects documentation)

That makes Projects suited to work that outlasts one reply and has more than one part: migrations across repositories, a research stream that needs several deliverables, or a build where implementation, tests, and review can proceed in parallel. Anthropic says you can configure the project's goal, repository or other context, cloud environment, connectors, plugins, instructions, and model. (Anthropic's announcement)

Repository Context and Shared Memory

Every new thread starts with the project's repositories and files, instructions and memory, connected tools, and cloud environment. The environment controls network access, environment variables, API credentials, and installed tools. Each repository is cloned for the thread, and its CLAUDE.md and skills are loaded. (Claude Code Projects documentation)

The project also has shared memory and a Library. Threads can draw on details accumulated by earlier work, while the Library collects files added to the project and files produced by threads. This is the part that turns a set of parallel runs into a continuing workstream rather than independent sessions. (Anthropic's announcement; Claude Code Projects documentation)

The boundary is important: a cloud thread cannot use files or tools that exist only on your machine. A local database, emulator, VPN-only API, or other machine-only service still requires a local session or an agent view. The documentation also says a local session cannot be part of a Project. (Claude Code Projects documentation)

Access and Limits

Projects are in public beta on Pro and Max plans and are rolling out gradually, starting with accounts that have used cloud sessions and do not have existing projects in Claude or Cowork. Team and Enterprise plans are not available yet. If Projects does not appear at claude.ai/code or in the Code tab of the desktop app, the rollout has not reached that account; Anthropic provides a waitlist for users who want to request earlier access. (Claude Code Projects documentation; Anthropic's announcement)

The beta is available through Claude Code on the web, the Code tab of the desktop app, and the Claude mobile apps. It is not available in the terminal CLI, and it cannot use a local session as one of a project's threads. (Claude Code Projects documentation)

Projects use the same plan limits as other Claude Code sessions, but they can consume those limits faster. Every running thread is a full session, and the project conversation uses tokens too. Threads that watch pull requests can use the plan again when a CI failure or review comment arrives. A thread that reaches its limit waits for the limit to reset; routine-started threads instead stop with a limit error. (Claude Code Projects documentation)

What This Changes for Developers

Projects make coordination the primary workflow. Instead of manually opening several sessions, copying context between them, and stitching results together, you can give Claude a stream of related work and let the coordinator decide when to fan out. You still need to choose the repository context carefully, watch for branch conflicts, and account for the fact that parallel sessions spend limits faster.

The practical promise is not simply “more agents.” It is a persistent work queue with shared memory, repository-aware threads, and a single place to review what is finished, what needs attention, and which pull requests are ready. For teams that need local files, private network services, or organization-level controls, the current beta's cloud-only and single-user boundaries are just as important as its parallelism.