Ambient Suggestions consumes many tokens on start up - Add warning

What version of the Codex App are you using (From “About Codex” dialog)?

26.519.81530 (3178)

What subscription do you have?

Plus

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

Ambient suggestions will fire off ChatGPT 5.4 requests to get your recent activity, including recent gmail or new tasks. However, this can consume many tokens

For one request 389,689 tokens. The full split was:
input_tokens: 385,039
cached_input_tokens: 312,064
non_cached_input_tokens: 72,975
output_tokens: 4,650
reasoning_output_tokens: 2,881
total_tokens: 389,689

This is about $0.63, if through API.

The next day, again (5) 5.4 requests with prompts like "Generate 0 to 3 hyperpersonalized suggestions for what this user can do with Codex in this local project:". The total token use was:
input_tokens: 991,298
cached_input_tokens: 802,048
non_cached_input_tokens: 189,250
output_tokens: 5,219
reasoning_output_tokens: 1,809
total_tokens: 996,517

This is about $0.75, if through API.

This can consume up to 11% of a 5 hour window on Plus, which is absurd enough to encourage CLI over the app.

What steps can reproduce the bug?

Just open the codex app after a while and navigate to a project/overaching project. The auto-suggestions will fire over all connectors in the project.

The data is in the .codex/logs_2.sqlite file. If you have not used 5.4 that day (easy to avoid when you have sent NO activity), the logs can be found in "logs."

What is the expected behavior?

Add a warning, and have it not be the default (if it currently is).
Seperately, in the future, Do it through new "batch" APIs and charge significantly less.

Additional information

_No response_