Fix Cowork Feature’s 10GB VM Bundle Performance Issues

Users of the Cowork feature in Claude Desktop are experiencing severe performance degradation due to a 10GB VM bundle that accumulates over time. This issue causes slow startup times, UI lag, and unresponsive behavior during sessions. Let’s explore the problem and how to resolve it.

Understanding the Problem

The Cowork feature generates a VM bundle file at ~/Library/Application Support/Claude/vm_bundles/claudevm.bundle/rootfs.img. This file grows to 10GB and remains uncleaned by the application. Even after deletion, it regenerates rapidly, causing persistent performance issues.

Key Symptoms

• Slow application startup
• UI lag during normal use
• Severe slowdowns during extended sessions
• Failed or hung tasks

Impact on Performance

Testing shows that deleting the vm_bundles, Cache, and Code Cache directories reduces disk usage from 11GB to 639MB. This cleanup results in approximately 75% faster performance on previously problematic tasks.

Performance Degradation Timeline

1. Initial use: 10GB VM bundle created
2. Performance issues emerge within hours
3. Issues worsen over multiple sessions
4. Complete system slowdown occurs

Steps to Resolve the Issue

Immediate Cleanup

1. Close Claude Desktop
2. Delete the vm_bundles directory
3. Clear Cache and Code Cache folders
4. Restart the application

Preventative Measures

• Disable the Cowork feature if not needed
• Manually clean VM bundles weekly
• Monitor disk usage regularly
• Report persistent issues to the development team

Preventing Future Issues

While the current solution requires manual cleanup, users can:

• Request automatic cleanup functionality
• Monitor for official patches
• Use alternative collaboration features

Conclusion

The Cowork feature’s 10GB VM bundle significantly impacts performance. By understanding the issue and implementing regular cleanup procedures, users can maintain optimal system performance. For developers, addressing this issue through automated cleanup mechanisms would provide long-term solutions.

Call to Action

Have you experienced similar issues? Share your solutions in the comments or follow this repository for updates on the Cowork feature’s performance improvements.