Your agent hits a usage limit.
It keeps working anyway.
When the Claude, Codex or Grok Build account an agent runs on is about to exhaust its quota, a quota rule hands the same conversation to another account you are already signed in to, and the work continues.
No restart, no lost context, no waiting for the 5-hour or weekly window to reset. It turns several smaller subscriptions into something that behaves like one much bigger plan.
One conversation, two accounts. The handover happens where the wall used to be.
Every Claude, Codex and Grok Build plan is capped: a rolling window measured in hours, and a longer one measured in days or a week. Hit the cap and the CLI stops mid-task. The agent was three files into a refactor, it had the whole plan in its head, and now it is frozen until a clock you cannot see runs out.
The usual fixes are all bad. You wait. Or you open a second account by hand, and start the conversation over from an empty prompt because the session lives inside the first account. Or you pay for the largest plan on the menu so the wall arrives later.
AgentsRoom takes the fourth path, and since version 1.188 it takes it through a quota rule. The Usage panel in the footer already reads every quota bar each CLI publishes; a rule written in that same panel says: when the session or the week of my most constrained account reaches 99%, move the running agents to another account. When it fires, AgentsRoom picks another account you are signed in to, moves the conversation transcript across, relaunches the session on that account, and tells the agent to pick the work back up where it stopped. From your side it looks like nothing happened, except a line in the terminal naming the account that took over.
A usage limit costs more than the wait
The wait is never just the wait. A five-hour window that opens at 3pm is the rest of your afternoon, and the task you were mid-way through is still unfinished when it reopens.
Restarting the conversation costs real money. Re-explaining the feature, the files, the decisions already taken burns tokens on the new account before it does a single useful thing.
Unattended runs die silently. A scheduled task at 2am, a long backlog ticket, a team workflow: they stop at the limit and you find out in the morning, with nothing done.
It happens repeatedly. This is not an edge case you hit once a quarter. For anyone running several agents in parallel, it is a weekly tax.
Several smaller plans, one uninterrupted day
The reason people ask for this feature is rarely convenience. It is the bill.
One large subscription
- : You pay the top tier to push the wall further away, not to remove it.
- : The cap is still there, just later in the day, and usually at the worst moment.
- : A single account means a single quota: two agents working in parallel burn it twice as fast.
- : When it runs out, everything stops at once. There is nothing to fall back on.
Two or three smaller plans
- : Each account carries its own independent quota and its own reset clock.
- : The first one to run out hands over to the next, mid-conversation, without you doing anything.
- : Total cost is often lower than the single top tier, for strictly more working hours.
- : An exhausted account is benched until its window resets, then rejoins the pool by itself.
AgentsRoom does not sell you the subscriptions and takes no cut of them. It simply stops the ones you already pay for from sitting idle while another is blocked.
What actually happens, step by step
Five steps, none of which need you once the rule is written.
The quota bar crosses the threshold you set
AgentsRoom polls the quota bars each CLI publishes, the same ones the Usage panel shows, and evaluates your rule against them: which account (the most constrained one, or a named one), which window (any gauge by default, or the session or the week), which threshold, 99% being the closest to the wall the editor offers. Nothing reads the terminal any more: a conversation that merely talks about limits, or a file that quotes one, cannot trigger anything, and a rule fires once per quota window, not every time the bar is polled.
The exhausted account is benched
It is marked unusable until its quota window resets, using the real reset time the provider publishes. An agent launched in the meantime is bound to a healthy account instead of walking into the same wall, and a second rule can hold or redirect scheduled launches before they even start.
A replacement is chosen
Among your other accounts for that provider, AgentsRoom keeps only those that are actually signed in on disk, not benched, and not already close to their own limit. It picks the one with the most quota left. Accounts you excluded are never considered. A quota rule can favour the next account to reset instead, so less quota goes to waste.
The conversation moves across
For Claude, the session transcript lives inside the outgoing account's own directory, so it is copied into the incoming account's tree first. Without that step the session would resume empty, which is exactly the loss the feature exists to prevent. The original account keeps its copy untouched.
The agent resumes and carries on
The CLI is relaunched on the new account, resuming the same session with its full history, and receives a short instruction to continue the interrupted work rather than restart it. A line in the terminal names the account that took over, and the agent's account badge updates everywhere in the app.
How each provider isolates its accounts
The switch is built on the mechanism each CLI already provides, not on a wrapper around it.
Claude Code
An account is a directory holding its own credentials, sessions and project history. Switching means relaunching the CLI against another directory, which is why the transcript has to travel with the conversation. AgentsRoom handles that copy for you.
CLAUDE_CONFIG_DIR
Codex
Codex isolates credentials the same way, but the conversation rollouts live in the per-project home AgentsRoom generates, which does not move when the bound account changes. The switch is therefore just a credential rebind and a resume.
CODEX_HOME
Antigravity CLI and Cursor
Both CLIs publish a quota bar since version 1.190 (Cursor's billing cycle, the 5-hour and weekly windows of each Antigravity model pool), but one reading for the whole machine, not one per account, and a rule that moves agents needs a bar per account: there is still no automatic switch for Antigravity or Cursor. Their accounts still exist and still run in parallel. An Antigravity account here is a Gemini API key, one per agent if you like, so several agents work on separate quotas; a Cursor account is its own configuration directory. When one of them runs out, you rebind the agent to another key or account by hand from its card, and the agent restarts cleanly rather than resuming the conversation.
GEMINI_API_KEY
Google AI Pro and Ultra subscriptions cannot be alternated: the Antigravity CLI keeps those logins in your operating system keyring, which holds one account for the whole machine and offers no way to pick between several.
This changes every Antigravity session on this machine, including the ones you start yourself outside AgentsRoom. They will use the API key instead of your Google subscription. Turn it off to go back.
What people do today instead
Each of these is a real workaround, and each of them costs something.
Wait for the window to reset
Free, and the most expensive option there is. The agent holds a plan it will lose to a context compaction, the branch stays half-done, and any unattended run scheduled in that window never happens at all.
Switch account by hand
Workable if you happen to be watching. You still lose the conversation: the session belongs to the account that created it, so the new one starts from an empty prompt and you re-explain everything you already explained.
A shell wrapper or an account switcher
These swap the active credentials globally, outside your editor. They cannot know that one particular agent, in one particular project, just hit a wall mid-turn, and they have no way to carry that conversation across. They also move every other session with them.
Buy the biggest plan
This moves the wall, it does not remove it, and it scales badly: running four agents in parallel divides the same quota four ways. Several independent quotas absorb a burst that one large quota cannot.
A work account will not silently pay for personal work
Automatic switching is off until you write the rule. Moving a conversation onto another subscription is a billing decision, and the one it most often concerns is a work account paying for personal work, so it is something you turn on rather than something you inherit. If you had enabled the former global switch, the update wrote the equivalent rules for you, named "Keep going when an account is exhausted", and you can read, disable or delete them like any other rule. Once a rule is on it is still never blind. Any account can be marked "Never used as backup", which is the right answer when you keep a strict separation between an employer's subscription and your own, or between two clients. Every switch is announced in the agent's terminal and the account each agent runs on is visible on its card in the sidebar, so what is being billed to whom is never a guess.
Do I need two subscriptions for this to do anything?
Does the agent lose the conversation when it switches?
Is it on by default?
Can I stop one specific account from being used as a backup?
What if every account is out of quota?
Could it switch by mistake and burn my second subscription?
Does the exhausted account come back?
Does this work for unattended runs?
Which providers support it?
Is this against the providers' terms?
Can I alternate two Google AI Pro or Ultra subscriptions on Antigravity?
What does putting an Antigravity agent on an API key change on my machine?
I added my Antigravity accounts and nothing changed. Why?
You may also like
Claude Multi-Account
Run two or more Claude Code accounts on one computer, side by side in AgentsRoom. Keep a personal account on one project, a work account on another, a client account on a third. Pin one account per project, override per agent, sign in from the app. Free to download on macOS, Windows and Linux.
Codex Multi-Account
Run several Codex CLI accounts side by side in AgentsRoom. Use a personal OpenAI account on one project, a work account on another, a client account on a third. Pin one Codex account per project, override per agent. Sign in directly from the app via codex login, no terminal hack required.
Token usage : per session : live
Live token meter on every Claude Code session in AgentsRoom. Red overuse alert, cache hit rate, full input, output and cache breakdown. 100% local. Free to download on macOS, Windows and Linux.
Multi-provider
Run Claude, Codex and the other agent CLIs side by side in the same project, and move a session from one to another without losing context.
Adaptive Mode
Right-size the model before you send: light tasks on a cheap model, heavy ones on a flagship. Fewer tokens spent means the limit arrives later.
Resume a session
Pick a previous CLI conversation back up instead of starting from an empty prompt, on any provider that supports it.
Scheduled tasks
Run agents on a schedule, including overnight. Combined with account auto-switch, a quota wall no longer ends the run.
Usage alerts
Before the wall: get a notification when a quota bar crosses 50%, 75%, 90% or a threshold you chose, on the desktop or on your phone, once per window.
Claude and Codex together
Before the wall: share new agents between Claude and Codex up front, one ticket out of two or a slice of each quota, so neither plan runs out alone.
Message queue
The other end of the same window: stack a prompt and arm it to fire when the quota comes back, or just before the window resets, so the credits you paid for get spent.
OmniRoute Alternative
Another answer to the same wall: a model gateway sends the request to another model, the account switch keeps the model and changes the account. The two approaches, compared.
Further reading
How Many Tokens Do I Have Left in Claude? The Two Screens to Check.
Your Claude plan is not denominated in tokens, so no screen shows a token balance. What it actually meters, the two screens that do show it, why it drains while you are not typing, and what the September 2026 weekly-window change does to your week.
Claude Code Keeps One Login at a Time. Here Is How to Run Several.
A field guide to running a work account and a personal account on the same machine: the one environment variable that decides which login is active, why the shell approach breaks once you have more than two terminals, and how to pin an account per project.
Antigravity CLI Keeps One Google Login Per Machine. Here Is What Works Instead.
Why you cannot alternate two Google AI Pro subscriptions on the Antigravity CLI, where it actually stores your login, what the account switchers really do to your system keyring, why a family plan does not double your quota, and the one approach that genuinely runs several accounts in parallel.
Stop losing afternoons to a usage limit
Sign a second account in once. The next time a quota runs out, you will find out from a line in the terminal, not from an agent that stopped hours ago.
Companion app: monitor your agents on the go
Bring your own: Claude, Codex, Antigravity CLI, or other AI provider.
Push bugs and requests straight to your public backlog.
Agents you read like a conversation
Each agent has a role, a face, and a thread you read like a messaging app.
