Anthropic Rebuilds Claude Projects With Parallel Threads and Shared Memory
A coordinator thread now manages parallel worker sessions - each a full Claude Code environment on its own repository branch
Claude Projects got a full rebuild. Anthropic scrapped the old folder-style container and replaced it with a thread-based architecture - a coordinator thread manages multiple worker threads running in parallel, each one a complete Claude Code cloud session operating on its own repository branch. Scoping, delegation, execution monitoring, and result assembly all happen without you directing each step.
How the Thread Architecture Works
A coordinator thread sits at the top of every session. Send Claude a task and it scopes the work, breaks it into sub-tasks, and routes each one to a separate worker thread - all without input from you. Workers run in parallel. Each runs as an independent cloud session inside Claude Code, isolated on its own repository branch with no visibility into what the other workers are doing at the same time.
Anthropic calls this Intelligent Work Delegation - a shift from you directing Claude through a workflow to Claude managing its own parallel sub-agents. You give a direction. Claude decides which threads handle which pieces, monitors each one as it runs, reviews the outputs, and assembles the final result. Building a system where Claude manages its own sub-agents rather than waiting for step-by-step instruction is the obvious next step for agentic tools - the surprise is that it required a ground-up rebuild rather than a feature addition.
Dynamic Memory Across All Threads
Memory now lives in a shared pool. Every worker thread inside a project can read from it and write back to it, which means context built during one task stays available in the next without you pasting background into every new session. Anthropic says the pool grows the more you use a project, reducing how much setup work each new conversation requires.
Projects you already created stay functional and will not break. Anthropic says existing projects will be auto-upgraded to the new thread-based architecture as the rollout expands. A waitlist is available now if you want early access before the automatic upgrade reaches your account.
Everything New in the Redesign
| Feature | What It Does | Status |
|---|---|---|
| Coordinator Thread | Scopes requests, delegates tasks, reviews outputs, assembles results | Beta |
| Worker Threads | Parallel Claude Code cloud sessions on separate repository branches | Beta |
| Dynamic Memory | Shared context pool across all threads, grows automatically over time | Beta |
| Project Library | Stores uploaded files and Claude-generated artifacts persistently | Beta |
| Usage Controls | Per-project monitoring, model selection, effort level for each thread type | Beta |
| Local Machine Execution | Run worker threads on your own hardware instead of Anthropic's cloud | Coming soon |
| Native Tools | Direct API calls, code execution, web browsing without manual config | Coming soon |
| Behind-Network Operation | Full project runs inside corporate infrastructure | Coming soon |
The Project Library Stores Everything
Projects now ship with a dedicated Library. Upload your own files - documentation, codebase context, brand guidelines, reference materials - and Claude stores the artifacts it generates in the same location. Anthropic is positioning Claude Projects as a persistent workspace rather than a session that resets, and the Library is the piece that makes that concrete.
Every worker thread pulls from this shared collection during a session. Generated reports, code outputs, processed data - all of it accumulates in one place you can browse, reference in new tasks, or export. Shared memory that persists across sessions is probably the most practical change here for teams running production workflows, since current Claude tools have forced developers to re-paste context at the start of every new conversation.
Who Gets Access First
Access starts narrow. Claude Pro and Max subscribers using cloud sessions in Claude Code get first access - the same subscriber tier that expanded into Claude Cowork earlier this year. One catch: this initial beta wave excludes anyone who already has existing projects. Anthropic says it plans to expand to more Claude Code users over the coming week.
- Claude Pro and Max on Claude Code (cloud sessions, no existing projects) - available now in beta
- More Claude Code users - rolling out over the next week
- Claude chat subscribers - follows the Claude Code expansion
- Enterprise subscribers - rollout comes after the chat expansion
Three Features Still on the Roadmap
Anthropic listed three capabilities still coming after the current beta. Local machine execution would let worker threads run on your own hardware rather than in Anthropic's cloud, opening the system to private environments and offline workflows. Native tools integration means workers will call external APIs, run code, and browse the web without manual configuration. Behind-network operation targets enterprise deployments where data must stay inside internal infrastructure.
Usage Controls - available now - already let you pick the model and effort level for both the coordinator and worker threads independently, which matters most for teams that reconfigured their defaults after Claude Sonnet 5 launched with agentic tool use earlier this year.
- Local machine execution: worker threads run on your hardware, not in Anthropic's cloud - opens the system to air-gapped and private environments
- Native tools: direct integration with APIs, code execution, and web browsing without manual setup per project
- Behind-network operation: designed for enterprises where all data must stay inside corporate infrastructure
Frequently Asked Questions
Do I need to manage the worker threads myself?
Will existing Claude Projects work with the new system?
Can I use the redesigned Claude Projects in Claude chat?
What goes in the Project Library?
What models run in each thread?