Usage and Limits

Usage and limits

phoenix usage reports normalized account limits for the active provider. /usage adds the current chat's input and output tokens and any catalog-based cost estimate.

Normalized states

• Ready means the reported allocation has comfortable capacity.
• Low means reported remaining capacity is near the configured threshold.
• Exhausted means the provider reports no usable allocation.
• Unknown means no reliable limit source is available.

Unknown is not converted to zero and normally fails open. Live provider errors can still activate bounded retry, key rotation, or model fallback.

Dedicated sources exist for supported Anthropic OAuth usage, ChatGPT Codex usage, OpenRouter credits and key budgets, DeepSeek balance, Google OAuth quota, and local Ollama availability. Data is cached briefly. Windows can apply to specific models, and multiple credentials can report separate pools.

During Colab, a seat strictly above 98 percent reported use can delegate its complete task assignment to the existing peer. Exactly 98 percent does not trigger delegation. Unknown usage does not sideline a seat. The saved model choices are unchanged.

Provider reporting is advisory and differs in coverage and freshness. See Providers and Colab.