[Desktop/Cowork] Repeated system-wide freezes on Windows 10 — memory exhaustion during sessions with the built-in browser (Windows flagged claude.exe with RADAR_PRE_LEAK_64)

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

On Windows 10 (16 GB RAM), long Chat/Cowork sessions in Claude Desktop — especially with the built-in browser pane doing web research — drive the whole system into memory exhaustion. The entire machine hard-freezes and needs a power cycle. This happened 7+ times between 15–28 Aug 2026 (Windows Event ID 41/6008 "previous system shutdown was unexpected" each time), and eventually led me to reinstall the app (which didn't help).

Evidence collected from the machine:

• Windows' resource-leak detector fired RADAR_PRE_LEAK_64 against claude.exe v1.30096.5.0 (15 Aug, Event 1001, Report Id 62e5066e-75de-4e8a-8017-bed989867954). The machine froze later that same night.
• On 26 Aug, the app's own main.log shows system free RAM collapsing 2389 MB → 759 MB in 4 minutes, immediately after a second renderer (the in-app browser pane) spawned. [CliGovernor] logged "memory pressure (warning)" at 21:59:55; the machine then wedged during a shutdown attempt at 22:24 and had to be power-cycled (Event 41 at next boot).
• [CliGovernor] memory pressure (warning/critical) recurs across many evenings (19, 20, 21, 26, 29 Aug). At critical pressure it repeatedly finds nothing it can evict ("would evict 0 idle session(s), 0 effective").
• The freezes on 26 and 27 Aug both occurred during the same Cowork session doing web research with the built-in browser pane.

Unexpected-shutdown log (System event log, all requiring power cycles): 16/08 ~00:15, 21/08 ~08:07, 21/08 ~23:51, 23/08 ~14:39, 26/08 ~21:56, 27/08 ~22:41.

What Should Happen?

Claude Desktop should bound its memory usage — or shed browser/renderer memory under pressure — so that a long research session cannot exhaust system RAM and freeze the entire machine. The memory governor detects critical pressure but has nothing it can evict; at minimum the app should warn the user and degrade gracefully before system-wide exhaustion on 16 GB machines.

Error Messages/Logs

--- %LOCALAPPDATA%\Claude\Logs\main.log, night of the 26 Aug freeze ---
21:56:48 [process-memory] trigger=interval tree_rss_sum=955MB ... sys_free=2389MB/16288MB
21:57:47 [process-memory] tree_rss_sum=1039MB ... sys_free=1873MB/16288MB
21:58:43 [process-memory] tree_rss_sum=1191MB electron(14) top=[electron_renderer:15224:431MB electron_main:16776:229MB electron_renderer:12332:205MB ...] sys_free=1355MB/16288MB
21:59:41 [process-memory] tree_rss_sum=1222MB ... sys_free=955MB/16288MB
21:59:55 [warn] [CliGovernor] memory pressure (warning): would evict 0 idle session(s), 1 effective
21:59:55 [process-memory] trigger=pressure tree_rss_sum=1226MB ... sys_free=759MB/16288MB
22:24:43 [info] Windows session ending (shutdown) - quitting the app   <-- log ends; machine wedged, power-cycled; Event 41 at next boot

--- Recurring governor warnings ---
2026-08-20 22:24:53 [warn] [CliGovernor] memory pressure (critical): would evict 0 idle session(s), 2 effective
2026-08-29 13:53:25 [warn] [CliGovernor] memory pressure (critical): evicted 0 of 0 idle session(s), 0 effective

--- Windows Application event log ---
15/08/2026 9:35:17 PM — Event 1001, Windows Error Reporting
Event Name: RADAR_PRE_LEAK_64
P1: claude.exe
P2: 1.30096.5.0
Report Id: 62e5066e-75de-4e8a-8017-bed989867954

--- Other app errors in the same period ---
2026-08-22 00:19:25 [error] Session ... query error: Claude Code process exited with code 1073807364 (0x40010004)
2026-08-28 22:33:56 [info] Windows session ending (critical) - quitting the app

Steps to Reproduce

Not deterministic, but reproduced 7+ times in two weeks on the same machine:

1. Windows 10 Home 19045, 16 GB RAM, Claude Desktop (MSIX install from the Microsoft Store path C:\Program Files\WindowsApps\Claude_...).
2. Open a Chat and Cowork session and do web-research-style work using the built-in browser pane (in my case: watch price research with multiple source lookups). Several MCP servers connected.
3. Leave the session running for an evening (1–3 hours), optionally with one or two other local sessions idle.
4. Watch [process-memory] lines in %LOCALAPPDATA%\Claude\Logs\main.log — sys_free trends down until [CliGovernor] memory-pressure warnings appear.
5. Eventually the whole machine hard-freezes (no BSOD — display frozen/unresponsive) and needs a power cycle. Windows records Event ID 41/6008 at next boot.

Claude Model

Opus

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

Claude Desktop 1.40609.0 (Windows MSIX); issue observed across 1.30096.5.0 → 1.32885.1.0 → 1.37937.3 → 1.40609.0

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Other

Additional Information

• This is the Claude Desktop app (Chat and Cowork) on Windows, not the CLI — filed here as the closest public tracker. Happy to be redirected if there's a better channel.
• Hardware: Gigabyte B550I AORUS PRO AX, 16 GB RAM.
• The Claude process tree itself was only ~1.0–1.2 GB RSS in the samples above, but total system free memory collapses in minutes while the app is the main foreground workload — suggesting memory growth in processes outside the tracked Electron tree (Cowork VM node / browser children) or a leak that RADAR caught on the main process.
• Reinstalling the app did not help; freezes continued on the new version.
• Sentry crash telemetry from this install may correlate: org 7ec346dd-575d-4e81-aae2-136220d99909, Cowork session local_f3832fce-7acc-4406-bbfc-a94423ac8229 (active during the 26 and 27 Aug freezes).
• Full WER reports and logs available on request.