Token Tunnel requests wait their turn instead of failing

When many chats and workflows hit the same hosted Token Tunnel subscription at once, for example a batch of scheduled workflows that all start in the same minute, the extra requests used to retry on their own schedule, back off for up to half a minute at a time, and give up after about a minute and a half. A workflow pinned to its subscription model could fail even though the subscription itself was fine. Now each request waits its turn in line and starts the moment a slot frees.

How it works

• First come, first served. A request that finds every slot in use joins a queue and is served in the order it arrived, as soon as a running request finishes.
• Capacity per provider. Claude, ChatGPT, Gemini, Grok, and Muse each get their own slots, so a busy Claude subscription never holds up a ChatGPT request.
• Room for bursts, and more with every account. Each connected account takes up to 50 requests at once, and a second Claude account doubles how much Claude work runs at once.
• Waiting is not slowness. Time spent in line no longer counts against the time we give a provider to start answering, and the chat shows "Waiting for a free subscription slot" while it waits.
• Auto still moves on. If a request waits a full minute, an Auto run switches to its next model, while a run pinned to the subscription keeps waiting.