13When a subscription runs out

Keep working
when quota runs out.

Sign in to one provider with several subscriptions and API keys. Auto keeps each conversation on its account while the prompt cache is warm and moves it before the quota runs out. When every subscription of a provider is spent, the conversation can move to another provider you list.

  1. 20%+StayA healthy account keeps the conversation and its warm cache.
  2. <20%Move when coldA low account hands the conversation to a healthy one at its next cold point.
  3. <5%Move nowA spent account hands it over at the next request.
  4. →Next providerWith every subscription spent, it moves to the next provider you listed.
IllustrationWhat Auto does at each quota level, with the default thresholds.How Auto picks an account

  • Moves that wait for a cold cache

    A provider caches the prompt per account, so a move sends the whole prompt again at the full input price. Auto keeps a conversation on its account and moves it at a cold point, such as after compaction or a long idle, once the account runs low. A spent account moves at the next request.

  • Two strategies

    ordered, the default, uses the first usable account in your priority order and goes back to it once its quota resets. paced compares the time passed with the quota used in each window and picks the account with the most room, so quota that is about to reset is used first.

  • API keys only as a fallback

    Auto keeps subscription conversations on subscriptions. It moves one onto an API key only when the key sets fallback = true and every subscription is spent, and moves it back at the next cold point.

  • Across providers

    List providers under [switching] in providers.toml. When every subscription of one is spent, a conversation moves to the next on a favorite model of the same size, such as a flagship for a flagship. It returns to your model once that provider has room again, and a run moves at most twice.

  • Your choice when it is spent

    When a pinned account, or every account on Auto, is spent, a card offers another subscription, an API key for this run, a wait until the reset, or another model.

  • Quota in view

    /accounts lists every account with the quota left in each window and when it resets. With two or more accounts, the status bar shows the account and the quota left in its tightest window, as [auto·personal 41% wk], and /usage keeps spend per account.

  • Every move in the transcript

    Each switch adds a row with its cause and the size of the cache it rebuilds, such as Switched Anthropic account personal → work: personal weekly 96% used. Prompt cache rebuilds (~84k tokens).

Claude and ChatGPT subscriptions report their quota with every response. Other providers have a reading when they report usage, and an account without one counts as healthy. Every move rebuilds the prompt cache on the new account. Claude subscription sign-in is experimental. Anthropic's terms limit Pro and Max subscriptions to official clients.

Accounts