Up to 93 days, inclusive in UTC. Totals cover the full filtered range, not just the visible page. LiteLLM's daily rollups may lag recent requests; spend is an estimate, not a provider invoice.

Request metadata recorded by LiteLLM. Prompts, responses, and credentials are not displayed. Models are recorded upstream names; provider-default changes do not rewrite history.

User limits are shared across all that user's keys; key limits apply in addition. Alerts compare current LiteLLM spend with each individual cap, not the Usage date range. Spend may lag requests and concurrent requests can exceed a cap. No external notifications are sent.

Choose where each provider's default alias sends new requests. Explicit model aliases are unchanged. Changes can affect cost and model capabilities.

Choose the chat model behind each PYTHIA alias. These independent routes do not change provider defaults. Saved targets survive restarts and apply to new requests; fast/smart do not assign models to particular tasks.

Developer access policy

Default limits are shared across each Workspace user's keys. Key and team limits may also apply; inspect a key for its assigned policy. Budget exhaustion still stops requests.

Model RPM/TPM are gateway deployment limits when configured; provider quotas are separate.