Honest limitations

The tradeoffs that come with pooled hot-swapping

  • One cold turn. Prompt cache is org-scoped, so the first turn on a new account re-uploads context once.
  • Depleted-pause hiccup. Only a fully depleted pool stops the session for the countdown; anything typed in that split second is lost.
  • Adoption lag. On macOS the first turn within ~30 seconds of a swap can still meter the old account; the bars' headroom absorbs it.
  • Shared blast radius. All default-profile sessions share one live credential, so a swap moves them all together (each adopts in place).
  • Plaintext credentials on Linux. Claude Code itself stores Linux credentials as a 0600 plaintext file; tokenmaxxing's parked copies follow the same model.
  • One pool per host. Pooling the same accounts on several machines works, but each host parks and rotates tokens independently, with no cross-host lock: two hosts refreshing the same account's token concurrently race, and for Codex a refresh from a stale parked copy is fatal to that login (OpenAI invalidates the whole grant family on refresh-token reuse, forcing auth on every host). If you pool an account on more than one machine, expect occasional re-auth churn on Claude and avoid it entirely for Codex.