Anthropic launched a new projects feature for Claude Code. The update lets customers launch multiple sessions related to a single project.
Users can access the tool from the Claude.ai sidebar. This feature differs from the older version of projects that provided self-contained workspaces with separate chat histories and knowledge bases.
The new implementation is currently in beta. Anthropic describes projects as a way to let Claude manage multiple related tasks.
Claude scopes the request and delegates the work. It coordinates parallel threads, reviews the outputs, and assembles the finished result.
Users can steer progress throughout the process even from a phone. The system keeps working after users step away from their computers.
This capability helps when prompting the model for a large coding project. Users can get components like authorization, databases, caching, and containers running simultaneously.
A project spins up each task as its own Claude Code cloud session thread. Each thread works on its own git branch and its own copy of the relevant repository.
A coordinator directs the threads. Overlaps surface as merge conflicts if threads touch the same code.
Subagents within each session may break threads down into distinct tasks. Claude Code now offers five different ways to run multiple tasks at the same time.
Projects are intended for work spanning multiple repositories or migrations. They also support non-code tasks like querying support tickets or documentation.
Anthropic warns that projects burn through tokens quickly. Each thread is a full Claude Code session, which causes users to reach usage limits faster.
Select Pro and Max subscribers have access to the Projects beta if they use Claude Code and lack existing web or desktop projects. Others can join a waitlist. Team and Enterprise plans lack access to the redesigned Projects beta for now.
