Quick answer
The local whoburnedmore CLI command is npx whoburnedmore --local. Decline its optional publish offer to keep the report local. The standard command signs in and submits daily aggregates. Neither mode submits prompts, source code, or filenames.
What this guide answers
The reader wants to know when to use local Codex versus cloud tasks.
Best for
Codex users choosing between quick local changes and larger delegated tasks.
Key facts, checked November 8, 2026
- Local messages and cloud chats share your plan's Codex usage allowance. OpenAI says cloud tasks may use more of that allowance than local messages. source
- On Plus and Standard Business, OpenAI estimates local messages per five-hour period by model: GPT-6.1 Sol 15-160, GPT-6.1 Terra 35-285, GPT-6 Luna 350-3,000 and GPT-6 Astra 5-45. Weekly limits may also apply. source
- Codex used with an API key is billed at API rates and has no cloud features: no Codex cloud, GitHub code review or Slack. source
- Fast mode draws on included subscription usage at 2.5x the Standard rate, and Ultrafast at 8x. With purchased credits the multipliers are 2x and 6x. source
Local is easier to interrupt
Local work happens near your editor and tests. When the agent drifts, you can stop, inspect, and narrow the next prompt.
- Best for small patches.
- Good for test failures.
- Easy to reset context.
Cloud needs tighter briefs
Cloud tasks are powerful when the brief is specific. They become expensive when the agent explores too much without a checkpoint.
- Attach acceptance criteria.
- Limit the file scope.
- Ask for a plan before broad edits.
Watch shared limits
OpenAI describes Codex usage as depending on plan and task complexity. The exact number of messages varies, so the safe habit is to measure windows and outcomes.
- Track local and cloud separately.
- Flag large cloud tasks.
- Compare task cost to shipped value.
Use a routing rule
Route fast, interactive work local. Route independent chores cloud. Route ambiguous architecture decisions to a human review before either agent starts burning.
- Local: debug and patch.
- Cloud: bounded chores.
- Human: unclear scope.
Why cloud tasks cost more per task
A cloud task runs in an isolated workspace built from a prepared environment. It usually works through more steps without you watching: setting up, exploring, editing, running tests and preparing a diff. Every step adds context that is resent on later turns. Locally, you see each turn and can stop a drifting run after one bad step. That is why OpenAI warns that cloud tasks may use more of the shared allowance. OpenAI's credit table lists rates per model, not per surface, so the extra cost comes from the extra work.
- Prepare the cloud environment once (dependencies installed, tests runnable) so each task does not burn turns on setup.
- Write the acceptance test into the task brief so the agent knows when to stop.
- Code reviews triggered through GitHub (tagging @Codex or automatic reviews) count as Code Review usage. Reviews you run locally count toward general limits.
A practical routing rule
Keep work local when you expect to correct course often, such as debugging, unfamiliar code, or anything where you would read each diff. Send work to the cloud when the task is bounded and verifiable, and you would only look at the final pull request: dependency bumps, test backfills, scoped refactors. Cloud tasks keep running while your computer sleeps, which is worth paying for only if the brief is tight enough that you will not rerun it.
Measuring each side
The usage dashboard at chatgpt.com/codex/settings/usage shows combined consumption and reset times, and /status in the Codex CLI shows what remains during a session. whoburnedmore reads Codex local session logs under ~/.codex/sessions. It counts your local work in daily totals, but it does not see cloud tasks, which run on OpenAI's infrastructure. If whoburnedmore's Codex numbers look low against the dashboard, the gap is likely cloud work.
Common questions
Do Codex cloud tasks and local CLI use the same limit?
Yes. Local messages and cloud chats share your plan's allowance, though cloud tasks may use more of it per task.
Can I use Codex cloud with an API key?
No. API-key Codex covers the CLI, SDK and IDE extension at API pricing. Cloud features need a ChatGPT plan.
Where do I see how much of my Codex limit is used?
On the usage dashboard at chatgpt.com/codex/settings/usage, or by running /status in an active Codex CLI session.