A hint or classification of the current task (e.g., "code_generation", "summarization", "general_chat").
Optional requestingThe ID of the GMI or agent instance making the request.
Optional personaThe ID of the active persona guiding the interaction.
Optional activeThe full active persona definition.
Optional userThe ID of the user initiating the request.
Optional userThe subscription tier of the user, which might affect model availability or preference.
Optional queryThe user's query or the primary input text for the task.
Optional languageThe target language for the model's output (e.g., "en", "es", "ja").
Optional optimizationA preference for optimizing model selection (e.g., for speed, cost, or quality). This can guide the router if multiple suitable models are found.
Optional requiredExplicitly required capabilities for the model (e.g., "tool_use", "vision_input", "json_mode"). The router must ensure the selected model supports all listed capabilities.
Optional preferredA list of preferred model IDs, if any. The router should try to use one of these if suitable.
Optional preferredA list of preferred provider IDs, if any.
Optional excludedA list of model IDs that should be excluded from consideration.
Optional maxMaximum acceptable cost per 1000 input tokens (in USD), if cost is a critical factor.
Optional maxMaximum acceptable cost per 1000 output tokens (in USD).
Optional userUser-provided API keys for specific providers, which might enable access to certain models.
Optional customAdditional custom parameters or context to aid routing decisions.
Optional policyContent policy tier governing this request. Drives model selection: safe/standard use default censored models, mature/private-adult route to uncensored models via PolicyAwareRouter.
Optional contentFiner-grained content intent hint. A mature session doing combat narration vs. romance has different model preferences.
Parameters provided to the model router to aid in its selection process. This context allows the router to make informed decisions.