External Publication
Visit Post

Codex credits are draining while idle / not actively used

OpenAI Developer Community May 9, 2026
Source

I may have found the source of the issue. I have turned off Settings GeneralSuggested prompts- Suggest what to do next by searching project files and connected apps and I don’t see usage drain anymore.

It could be just coincidence but I believe this is the cause of the usage drain.

I checked ~/.codex directory and saw an ambient-suggestions directory with json files containing “suggestions” - And they look like prompt responses that I have not asked for. Then I used Gemini to track all codex logs while I changed to a project and prompted it and apparently Codex sent 88k tokens after prompt response while idle. It was sending git and file system info in prompts using gpt-5.4 model. I have 5.5 set. I can see it in the logs

codex-desktop-ccc-ccc-ccc-ccc-ccc-ccc-t1-i1-ccc-0.log:2026-05-09T01:11:28.547Z warning [git] git.command.complete aborted=true allowedNonZeroExitCodes=null argsCount=7 command=“git log --max-count=1 ‘–since=7 days ago’ '–author=xx commandArgs=[“git”,“log”,”–max-count=1",“–since=7 days ago”,“–author=xxx”,“–format=%H”,“–”,“.”] cwd=/Users/xxx durationMs=106 exitCode=0 failureReason=aborted hostId=local id=b391d86f isRemote=false outputLimitExceeded=false outputLimitMaxBytes=null outputLimitStream=null requestKind=has-recent-user-commits source=home_ambient_suggestions startTimeMs=1778289088441 stderrBytes=0 stdoutBytes=41 subcommand=log success=false timedOut=false
codex-desktop-eeee-eee-eeee-eee-eeee-eeee-t0-i1-eee-0.log:2026-05-09T02:45:12.648Z info [ephemeral-generation] ephemeral_generation_token_usage cachedInputTokens=41856 event=ephemeral_generation_token_usage feature=ambient_suggestions inputTokens=44314 model=gpt-5.4 outputTokens=663 reasoningEffort=medium reasoningOutputTokens=464 serviceTier=null status=success totalTokens=44977

Discussion in the ATmosphere

Loading comments...