Claude Code Projects beta lets one conversation coordinate parallel cloud threads
Anthropic's redesigned Claude Code Projects coordinates parallel cloud sessions with shared memory and mobile steering, now in beta for select Pro and Max users.
On 17 September 2026, Anthropic opened a beta of redesigned Claude Code Projects to select Pro and Max subscribers. The headline feature is straightforward: a single project conversation now acts as a coordinator, breaking a goal into separate threads and handing each one off to a full Claude Code cloud session running in parallel. Anthropic’s own framing for the announcement is “from folder to conversation,” which accurately captures the shift.
What changed and why it matters
The original Claude Projects, introduced in June 2024, was a workspace for grouping chats, files, and shared instructions. Each chat remained its own conversation, and carrying context between them required deliberate prompt engineering or manual copying into Project knowledge.
The redesigned Claude Code Projects solves a different problem: coordinating agentic software development across multiple sessions simultaneously. You describe a goal, and the coordinator breaks it into tasks that can run in parallel. Each thread gets its own Claude Code cloud session, its own branch, and its own copy of the repository. When two threads touch the same code, the overlap is resolved as a standard merge conflict rather than silently combined.
The coordinator runs Opus at high effort by default. There is no fixed thread cap below 200 new threads per day across a subscriber’s projects, and the number of concurrent threads is not hard-capped either. If you ask Claude to limit itself to three threads at once, that preference is saved and respected, but it is a preference, not an enforced ceiling.
Shared memory and the project library
Every thread adds to and draws from a shared memory store. Over time, Claude builds up knowledge about your project’s conventions, decisions, and architecture without you needing to re-explain them in each session. Alongside memory, projects now include a library that collects files you add and artefacts Claude produces across all threads.
Claude also remembers your preferred working style. You can instruct it to check in less often, spin up threads more aggressively, or keep updates brief, and those preferences persist across sessions.
Mobile steering
An Overview panel surfaces which threads are waiting for input. Mobile access, available through the Claude apps, lets you answer a question or redirect a thread from your phone without needing to sit at a desk. For longer-running builds, that matters: threads keep working after you close your laptop, and you retain visibility and control without being tethered to a desktop client.
What does this mean for you?
If you are a Pro or Max subscriber who uses Claude Code cloud sessions, the practical effect is that you can delegate a larger chunk of work in a single instruction and let Claude figure out which parts can proceed simultaneously. Deprecating an old API endpoint across three repositories, for example, can now be handled as one project goal rather than three separate sessions you coordinate yourself.
The honest caveat is one Anthropic states plainly in its support documentation: each thread is a full Claude Code session, and several running at once consume plan capacity faster. This is not free parallelism. It is an interface for spending a finite usage allowance more quickly, in exchange for having Claude manage the orchestration rather than you doing it manually. If a thread hits your plan limit mid-run, it waits and resumes when the limit resets. The exception is threads started by a routine: those stop with an error, and you need to restart them manually after the reset.
Cross-repository work is supported. A single project can simultaneously connect to an API repository, a web frontend repository, and a mobile app repository, which removes one of the more painful coordination tasks in multi-service development.
Availability and rollout
The beta is open from 17 September 2026 to select Pro and Max subscribers who already use Claude Code cloud sessions and have no existing projects in chat or in Cowork. Access widens to other Claude Code users on those plans in the following week, with Team and Enterprise plans coming after that. Anyone who does not see the Projects entry in the sidebar can join the waitlist. Projects live at claude.ai/code and in the desktop and mobile apps. They are not available in the terminal or through Amazon Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry.
Existing projects on Pro and Max plans continue working as before. Anthropic will upgrade them as the rollout expands.
A few technical details worth noting
Claude Code can now read agents.md, an instruction file format used by several other coding agents. By default it is used when a project has no CLAUDE.md of its own. If both files exist, CLAUDE.md takes priority unless you configure Claude to load both.
Cloud-only execution is the current constraint. Anthropic has indicated that support for running threads on a user’s own machine, alongside local tools and resources behind a private network, is on the roadmap, but that capability is not available yet.
The competitive field here includes OpenAI Codex, GitHub’s branch-based agents, and Google’s Jules, each of which approaches repository-level coordination differently. Anthropic’s specific choice is an explicit coordinator above the worker threads, with one conversation as the top-level interface and shared memory spanning all of them. Whether that architecture proves more natural to work with than alternatives will become clearer as the beta widens.