Guides
Checking usage
Your key is the only credential Korlogic needs, so there is no dashboard to sign into. Paste the key into the usage checker and it shows what is left and what you have spent.
The usage checker (permalink)
Open the checker and paste your key. The figures stay live, refreshing every five seconds, so you can watch a long job draw down a balance in real time.
Open the usage checkerNo account, no password. Just the key you already have.When a credit key runs low, top the same key up on the Top up page. Credit lands on the key you already hold, so nothing about your integration changes and no account is required. Token keys are renewed by whoever issued them rather than topped up here.
What it shows (permalink)
| Section | What you get |
|---|---|
| Key | Its label, billing mode, active or suspended status, and expiry date if it has one. |
| Credit | Credit keys only: granted, spent, remaining, and the percentage used. |
| Tokens | Token keys only: budget, tokens consumed, tokens remaining, and the percentage used. |
| Requests | Total requests recorded on the key, and how many were made today. |
| Recent activity | The latest requests with model, token counts, cost and outcome. |
| Allowed models | The key’s allowlist, or a note that every model is available. This is the authoritative answer: the /models endpoint returns the whole catalogue rather than your subset. |
How usage is counted (permalink)
- Credit keys are charged in US dollars, priced per model from the input and output tokens the provider reports. The same conversation costs different amounts on different models.
- Token keys spend actual total tokens, input plus output, against a fixed budget. The model makes no difference to how fast the budget drains.
- Settlement happens when the response completes. For a streamed response that means after the final event has flushed, so a long generation lands in the checker a moment after your last token.
- Failures are recorded at zero cost. A request the provider rejects appears as an error and is not billed. A request the gateway itself refuses, such as an unknown key or an exhausted budget, stops before the ledger and does not appear at all.
- Successful metadata calls are not recorded. A
GET /modelsreports no token usage, so it is neither billed nor counted as a request. Only calls that report usage reach the ledger.
The full breakdown of which outcomes are billed, and which appear here, is on Errors.
Per-request accounting (permalink)
To account for spend inside your own application, read the usage block on each response rather than polling the checker. It is the same count the gateway bills from.
{
"usage": {
"prompt_tokens": 412,
"completion_tokens": 168,
"total_tokens": 580
}
}The Anthropic Messages format reports the same information as input_tokens and output_tokens, with cache counts folded into the input side. For streamed OpenAI requests the usage block arrives on the final event, and you do not have to ask for it, as Quickstart explains.
What is recorded (permalink)
Per request: the model, token counts, cost, latency and HTTP status. A failure is marked with its status code only. Prompts, completions and the provider’s error body are never stored or logged.
Your key is never stored in a readable form. It is kept as a hash, which is why a lost key has to be replaced rather than looked up. See Authentication.