AI Router usage, balance and limits · Galaxium Hosting help

Track what you spend, what happens when your balance runs out, and the limits that can make a request fail.

See your usage

AI Router → Usage shows credits, requests and tokens for the last 24 hours, 7 days or 30 days. Times are in UTC.

• By model and By API key show where your credits go. Playground requests are listed as Playground.
• Recent requests lists single requests with their model, key, tokens, cost and status. Filter by key, model or status, for example Succeeded or Cached.

Request logs never contain your prompts or the model's answers.

See your balance

Your credit balance is in the top bar and on the AI Router overview, with an estimate of how many days it lasts at your 7-day pace.

Usage is measured per request to a tiny fraction of a credit and taken from your balance one whole credit at a time. Unsettled usage on Billing & credits is usage that hasn't added up to a whole credit yet.

Every API response also tells you what it cost: the x-galaxium-credits header (and the x_galaxium field in the response body) holds the credits charged, and x-galaxium-request-id the request number shown in Recent requests.

When your balance runs out

When your balance reaches 0, new requests fail with 402 (insufficient_credits) and the message "Your AI Router credit balance is empty. Top up to continue." Nothing else is affected, and requests work again as soon as you buy credits.

Before a request runs, the router sets aside the most it could cost. If that is more than your balance, you get a 402 even though you still have some credits. The message says how much the request could cost. Lower max_tokens or top up.

Tip: Turn on auto top-up under Settings so your apps never hit an empty balance, and set a Low-balance alert to get a warning in the portal first.

Limits

| Limit | What happens | |---|---| | Requests running at the same time | Too many at once on your account or on one key returns 429 (too_many_concurrent_requests). The message gives the limit. Retry when one finishes. | | A key's Monthly credit cap | Once the key has spent its cap this month (UTC), it gets 429 (key_monthly_cap_reached). Raise the cap under Edit limits, or wait for the new month. | | Model not allowed for a key | 403 (model_not_allowed). See Create and manage AI Router API keys. | | Request too big for the model | 400 (context_length_exceeded). Shorten the conversation or pick a model with a larger context. | | Request shape | Up to 500 messages per request, max_tokens up to 200,000, temperature from 0 to 2, n from 1 to 16. |

If you don't send max_tokens, a default cap is applied so the request can't run away with your balance.

Request log retention

Under Settings → Data retention, Keep request logs for sets how long single requests stay in Recent requests: 7, 30, 90, 180 or 365 days. Daily totals for charts, caps and billing are always kept.

Delete all request logs in the Danger zone clears the per-request history. Charts and billing are not affected.