Anthropic has widened its Claude Code Projects beta in a November 13 update, so coding agents can finally work on your own machine. Threads that once ran only in the cloud can now run locally through Remote Control, and the beta now covers Ultra subscribers alongside Pro and Max.
What changes for Projects
Projects turn one conversation into a coordinator for parallel coding sessions. Claude scopes each task, starts a thread for it, and assembles the results, with every thread running as a full Claude Code session on its own branch and its own copy of the repository.
The big change is local work. When a task needs a local database, a device emulator or an API behind a VPN, you can ask Claude to run that thread on your computer instead of the cloud. The thread then uses the machine’s own files, tools and permission rules.
Local threads have limits. They run only while the computer is awake with Remote Control on, they start without the project’s memory files loaded, and they cannot run at all while the Require trusted devices setting is on for your account.
Anthropic has also raised the ceiling on parallel work. The enforced limit is now 500 new threads a day across a user’s projects, up from 200, and the beta now includes Ultra plans. Team and Enterprise plans still wait for access.
How the coordinator works
Every project is one long conversation where Claude acts as coordinator. It takes what you send, decides what becomes a thread, and keeps track of each one. Quick questions get answered in place instead of starting a thread for them.
New projects run Opus everywhere by default, with high effort for worker threads and low effort for the coordinator conversation. Threads open pull requests on their own branches and keep watching them, pushing fixes when checks fail or reviewers comment.
Projects draw on the same plan limits as other Claude Code sessions, and they spend them faster. A thread that stops at a limit waits and then continues on its own when the limit resets, unless you switch that behaviour off.
Why it matters
Local threads close the biggest gap in the Projects beta. Since launch, cloud-only threads could not touch anything behind a firewall or on a developer’s disk, which kept real-world projects with private services out of reach for many teams entirely.
It also sharpens the contest with rival coding agents. Anthropic’s self-hosted agents already run inside company walls, and GitHub’s Copilot Projects span repositories — now Claude Code’s coordinator can follow work onto the developer’s own machine too, as well as the cloud.
Our read
This is a practical upgrade, not a flashy one. The coordinator idea only works if threads can reach the places real code lives, and local execution was the missing piece. Anthropic is fixing the beta’s rough edges before widening it, which is the right order.
The cost warning still stands. Every thread is a full session, so parallel work burns through plan limits fast, and the 500-a-day ceiling is a guardrail, not a target. Teams should set thread limits in project instructions before letting a coordinator run loose.
Background
Anthropic redesigned Projects in September, turning a static folder of chats and files into a conversation that manages parallel cloud sessions. The beta opened to select Pro and Max subscribers who used cloud sessions, with a waitlist for the rest.
Since then Anthropic has kept expanding Claude’s workplace reach, from Meetings in Google Meet to an office suite and self-hosted agents. Projects is the coordination layer underneath that push, and local threads make it usable for the code that matters most.


0 Comments