[Windows Desktop] Background ambient/home suggestions consume user quota without explicit action #29380

Summary

Codex Windows Desktop is consuming my ChatGPT/Codex 5-hour quota without explicit user action. Local logs show background turns triggered by ambient/home suggestions and their safety filtering shortly after opening Codex, before I submit a real coding request.

This is causing substantial quota loss. In one session, my 5-hour quota dropped from 100% to 97% within a few minutes after opening Codex. Across the morning, a significant portion of quota was consumed either by these background requests or by attempts to diagnose/mitigate the quota drain.

Product / Environment

• Product: Codex Desktop on Windows
• OS: Windows x64
• Codex package path observed locally: OpenAI.Codex_26.611.8604.0_x64__2p2nqsd0c76g0
• Auth mode in logs: ChatGPT plan
• Date observed: 2026-06-22

Evidence from local logs

Local logs_2.sqlite shows background model turns after opening Codex:

op.dispatch.user_input
model=gpt-5.4
cwd=<local project path>

The generated assistant output contained suggestions JSON:

{"suggestions":[...]}

A second background turn then ran the ambient suggestions safety/compliance filter:

You are an expert at upholding safety and compliance standards for Codex ambient suggestions.
model=gpt-5.4-mini
cwd=<Codex app install path>

Earlier logs also showed prompts like:

Generate 0 to 3 hyperpersonalized suggestions...
Codex ambient suggestions...

The background request is not just telemetry or websocket keepalive: it creates model responses and produces suggestion candidates.

Local mitigation attempts

I tried disabling memories and related local settings:

[features]
memories = false
js_repl = false

[memories]
generate_memories = false
use_memories = false

[desktop]
keepRemoteControlAwakeWhilePluggedIn = false

I also tried blocking writes to the local ambient-suggestions cache and setting cached generatedAtMs values into the future. This does not stop quota usage because the model request happens before the cache write.

I also tried adding best-effort config keys such as:

[features]
auto_review = false
home_suggestions = false
personalized_suggestions = false
ambient_suggestions = false
tool_suggestions = false
tool_suggest = false

There does not appear to be a documented, reliable user-facing setting to disable these background model calls.

Expected behavior

• Opening Codex should not consume user quota unless the user explicitly submits a request or opts into background suggestions.
• Ambient/home suggestions should either not count against quota or should have a clear setting to disable them.
• The desktop app should expose a switch such as "Disable background model calls / suggestions".

Actual behavior

• Codex Desktop automatically triggers ambient/home suggestion generation after launch.
• These requests use models such as gpt-5.4 and gpt-5.4-mini.
• The user's 5-hour quota decreases even when the user is not actively asking Codex to perform a task.

Impact

This creates a trust and billing/quota issue: paid users can lose quota simply by opening Codex, and additional quota is consumed trying to diagnose why the quota is being drained.

Please provide a supported way to disable all background model calls and ensure ambient/home suggestions are excluded from user quota unless explicitly requested.