The “Accounts” section keeps track of your LLM provider accounts and watches their quotas and balances while your agents work. When an account runs out of quota or money, you hear it from MultiAgent — not from crashing chats.
Account tracking
Supported providers are Z.ai, BigModel (Z.ai’s Chinese address), Codex, DeepSeek and OpenRouter. Each account records a provider, credentials, a status, colored tags and filters for them — so you can group “team A’s work accounts” or “the overnight reserve”. Counters at the top show how many accounts are at risk and how many are being watched.
Quota and balance monitoring
While you get on with your work, every enabled account is checked in the background on its own schedule — by default once every five minutes, and every check, successful or not, is stored as its own snapshot. The history lives in two layers: fresh snapshots give precision over the last two weeks, and hourly rollups live for years, so a long balance chart builds quickly without losing the picture of past months. That feeds the per-account analytics too: balance and quota over time, cuts by provider and by account, distributions, averages by hour and by day of week, and a snapshot table. The summary line shows balances per currency (from each account’s latest check), how many accounts are watched and how many are at risk — meaning 80% or more of the quota spent, or the provider reporting a limit reached. The moment a new check lands, charts and counters refresh themselves, with no reload. Account credentials are never shown in the report.
Signing in to Codex
Codex accounts sign in through ChatGPT right from MultiAgent: the “Sign in with ChatGPT” button starts the sign-in and opens the OpenAI page with the confirmation code — all you do is confirm. The status is visible on the page, and you can cancel a sign-in in progress or sign out of the account in the same place. Login data is stored encrypted and never returned to the browser; before the session expires, MultiAgent refreshes it on its own, and when the provider declines the refresh, the account honestly says it needs a new sign-in.
Codex quotas
Codex reports two limit windows: 5 hours and weekly. MultiAgent shows how full each window is and when it resets next, and warns when a reset is close. When you plan a heavy task, you can see whether it fits the window.
Models and accounts
Provider accounts and models are different things. An account exists so you can see a provider’s balance and quotas: nothing is passed from it into agents, and an agent’s keys live separately from the account. A model is a set of settings for a particular provider (name, icon, variables) that you pick in an agent’s chat. The model catalog also holds scheduling settings — how many chats to start at once and the pause between sends — but for now they are only stored and edited: nothing limits new chats by them.