Windows Codex 26.810.4967.0 renderer memory spikes and causes system-wide stuttering as a chat grows

Environment:

• Codex Windows app: 26.810.4967.0
• Windows 10 Pro, build 19045
• Intel i7-8750H, 12 logical processors
• 32 GB RAM

Observed behavior:
After using almost any chat for a while, the Codex UI and eventually the whole
computer become noticeably laggy. The problem appears to get worse as messages
and tool outputs accumulate.

Local measurements:

• One ChatGPT.exe renderer process reached about 2.15 GB working set.
• Total Codex-related memory approached 4 GB during a large tool-output render.
• After the output stopped/collapsed, the renderer dropped to about 754 MB and
  total Codex memory dropped to about 2.2 GB.
• CPU was not continuously saturated and the system still had about 18 GB free RAM.
• The current session transcript was only about 0.29 MB, so the issue does not
  appear to require a very large transcript file.
• GPU process was also active during the slowdown.
• ~/.codex/logs_2.sqlite is about 439 MB with roughly 186,000 log rows.

Expected behavior:
Long-running chats and large tool outputs should not cause renderer memory to
grow to multiple gigabytes or make the entire Windows desktop stutter.

Reproduction:

1. Start a local Codex chat.
2. Continue using it for tool-heavy work, especially commands with sizable output.
3. Observe ChatGPT.exe renderer memory and UI responsiveness.
4. Repeat with a new chat; the behavior eventually returns.

Reproduce the slowdown in the Windows Codex app with a tool-heavy chat while monitoring ChatGPT.exe renderer memory, GPU activity, and desktop responsiveness. Inspect ~/.codex/logs_2.sqlite alongside the renderer behavior, then verify that long-running chats and large tool outputs no longer grow memory to multiple gigabytes or cause system-wide stuttering.