Generation

Open Settings → Generation to configure default request parameters. Conversation-specific overrides may replace these values.

Common parameters

• Temperature controls randomness.
• Top P limits cumulative probability mass.
• Top K limits candidate tokens where supported.
• Maximum output tokens caps the generated response.
• Frequency and presence penalties discourage repetition where supported.

Agora sends only parameters supported by the selected provider protocol.

Thinking

Enable a thinking budget and choose either a token budget or a supported reasoning level: Minimal, Low, Medium, High, xHigh, or Max. Provider and model support varies; unsupported controls may be ignored or translated by that provider adapter.

OpenAI-compatible service tier

For compatible endpoints, the service tier can be Off, Auto, Default, Flex, or Fast. Availability, latency, and billing are controlled by the provider.

Related settings

Context budgeting and Compact are configured separately under Context. Automatic titles use Title Generation, and image-specific defaults use Image Generation.