How do I manage LLM API keys, usage, billing, and data security?
Learn how Lumoswitch uses separate access keys, model configuration, rate limits, call logs, and account billing to establish manageable security and cost boundaries for LLM API calls.
Maintained by the 路墨 Lumoswitch product team
Separate access keys, model configurations, and rate controls, together with account balance, call logs, and usage bills, create clearer permission and cost boundaries for LLM calls.
Why create a separate API key for each application?
A separate access key associates calling identity and model configuration with a particular application or user, making independent rotation, revocation, and investigation possible. Available models and call rates are determined by its model configuration, while the account boundary carries balance and billing. When unrelated applications share one key, a leak or anomalous call is difficult to locate and contain accurately.
How can I control which models an API key can call?
An access key is attached to a model configuration instead of maintaining another independent model allowlist. An administrator uses that configuration to select candidate resources, publish public model IDs, and set call rates. Use separate configurations and their respective access keys when model scope differs. Upstream resources and credentials remain under higher-permission administrator configuration.
Do call logs store full prompts and model replies?
Ordinary call logs record runtime metadata such as request time, model, status, latency, tokens, usage, and cost by default; they do not keep full prompts and response bodies as ordinary log content. For data-processing and retention boundaries, consult the current privacy policy, deployment configuration, and feature documentation.
How should I understand token usage, balance, and billing?
Platform-model calls create usage records from configured input and output prices and actual token usage. Account balance, promotional credit, reserved amounts, and overdue controls together affect the service boundary. The charging order, promotional amount, and price are determined by current console displays and applicable rules.
Will Lumoswitch send model requests outside the country?
Whether a cross-border transfer occurs depends on the model provider and processing region you select. Requests may be sent outside the country when you actively connect a provider located abroad or one that may process data abroad. To avoid cross-border processing, choose providers and regions that explicitly process data domestically and review their applicable terms.
What should I do first when I find anomalous usage?
First suspend or revoke the suspected leaked access key and inspect its recent call logs, model configuration, and rate settings. Then rotate a new key for affected applications rather than continuing to use possibly exposed credentials. For balance or billing disputes, retain call and bill records for the relevant period for reconciliation.