[BUG] Co-work VM causes daily system-wide memory exhaustion - crashes all Electron apps

Bug Report: Co-work VM causes daily system-wide memory exhaustion and app crashes

Summary

The Claude desktop app's co-work VM (com.apple.Virtualization.Virtual) causes daily memory exhaustion events that trigger macOS jetsam to kill processes system-wide, crashing Chrome, VS Code, Claude desktop, and other applications. This has occurred every day for at least 8 consecutive days.

Environment

• Hardware: Mac mini (Mac16,11), Apple M4 Pro, 12 cores, 24 GB RAM
• macOS: 26.2 (Build 25C56)
• Claude Desktop Version: 1.1.1890
• VM rootfs origin: fc96f5bc6a7aeee6e651437eba3c76d0fe9a16f3

Reproduction

1. Start Claude desktop app
2. Enable co-work with access to a local folder
3. Let co-work run for extended periods
4. System eventually hits memory pressure, triggering jetsam

Evidence: Daily JetsamEvents

macOS logs a JetsamEvent when memory pressure forces process termination. The Claude VM is the largest process in 7 of 8 events:

| Date | Largest Process | Compressor Size | Free Pages |
|------------|----------------------------------------|-----------------|------------|
| Jan 28 | com.apple.Virtualization.Virtual | 543,918 pages | 82,530 |
| Jan 29 | WindowServer | 81,102 pages | 7,283 |
| Jan 30 | com.apple.Virtualization.Virtual | 290,638 pages | 98,813 |
| Jan 31 | com.apple.Virtualization.Virtual | 469,852 pages | 89,275 |
| Feb 1 | com.apple.Virtualization.Virtual | 482,178 pages | 83,655 |
| Feb 2 | com.apple.Virtualization.Virtual | 32,942 pages | 407,265 |
| Feb 3 | com.apple.Virtualization.Virtual | 411,681 pages | 70,480 |
| Feb 4 | com.apple.Virtualization.Virtual | 609,674 pages | 88,320 |

Example from Feb 4 JetsamEvent:

• VM resident memory: 2.35 GB
• System compressor: 9.3 GB (609,674 pages × 16KB)
• System free: 1.4 GB (88,320 pages × 16KB)

Impact

When jetsam triggers:

1. Claude desktop: Goes to black screen, does not recover, requires force quit and restart
2. Co-work session: Completely lost with no recovery option
3. Chrome: All helper processes killed (SIGKILL exit code 9), app unresponsive
4. VS Code: All helper processes killed, app unresponsive
5. Other Electron apps: Similarly affected

Example from VS Code logs at crash time:

[error] Extension host with pid 65969: crashed with code 9 and reason 'killed'
[error] UtilityProcess fileWatcher: crashed with code 9 and reason 'killed'
[error] CodeWindow: renderer process gone (reason: killed, code: 9)
[error] ptyHost terminated unexpectedly with code 9

Why Anthropic isn't receiving crash reports

1. SIGKILL doesn't generate crash dumps - processes are terminated instantly with no opportunity to write diagnostics
2. Crashpad folders are empty:

• /Users/rtp/Library/Application Support/Claude/Crashpad/completed/ - empty
• /Users/rtp/Library/Application Support/Claude/Crashpad/pending/ - empty

1. The VM is a black box - no telemetry escapes from the VM before termination

Requested fixes

1. Memory management: Implement memory limits or monitoring in the co-work VM to prevent runaway growth
2. Graceful degradation: When approaching memory limits, warn the user and/or pause operations
3. Crash recovery: Claude desktop should recover from renderer process termination like Chrome and VS Code do, not show a black screen
4. Session persistence: Co-work sessions should be recoverable after crashes
5. Telemetry: Add monitoring that can detect and report memory pressure situations before SIGKILL

Attached files

The following JetsamEvent files document each occurrence:

• /Library/Logs/DiagnosticReports/JetsamEvent-2026-02-04-212444.ips
• /Library/Logs/DiagnosticReports/Retired/JetsamEvent-2026-02-03-211645.ips
• /Library/Logs/DiagnosticReports/Retired/JetsamEvent-2026-02-02-210130.ips
• /Library/Logs/DiagnosticReports/Retired/JetsamEvent-2026-02-01-222211.ips
• /Library/Logs/DiagnosticReports/Retired/JetsamEvent-2026-01-31-211852.ips
• /Library/Logs/DiagnosticReports/Retired/JetsamEvent-2026-01-30-201355.ips
• /Library/Logs/DiagnosticReports/Retired/JetsamEvent-2026-01-29-212547.ips
• /Library/Logs/DiagnosticReports/Retired/JetsamEvent-2026-01-28-213830.ips

Workaround

The only current workaround is to avoid using co-work with local folder access, or to frequently restart the Claude desktop app to reset VM memory state.