Struct ModelRouter 

pub struct ModelRouter { /* private fields */ }

Smart model router that picks cheap / mid / expensive models based on message complexity.

Implementations

impl ModelRouter

pub fn new(config: ModelRoutingConfig) -> Self

Create a new router from the routing configuration.

pub fn is_enabled(&self) -> bool

Returns true if model routing is enabled.

pub fn classify(message: &str) -> ModelTier

Classify a user message into a complexity tier using keyword heuristics.

The classification is intentionally simple: pattern matching on the lowercased message. No LLM call is made.

pub fn classify_with_context(message: &str, messages: &[Message]) -> ModelTier

Classify a user message with context awareness: if the conversation contains images (from screenshots, Telegram photos, etc.), force the expensive tier so a vision-capable model handles them.

pub fn select_model(&self, tier: ModelTier) -> Option<&ModelConfig>

Select the model config for a given tier. Returns None if the tier has no model configured (caller should fall back to the default model).

pub fn route_message(&self, message: &str) -> Option<(ModelTier, ModelConfig)>

Classify a message and create a driver for the selected tier.

Returns Some((tier, driver)) if routing is enabled and a tier-specific model is configured. Returns None if routing is disabled or the tier model is not configured (the caller should use the default driver).

pub fn route_message_with_context( &self, message: &str, messages: &[Message], ) -> Option<(ModelTier, ModelConfig)>

Classify a message with conversation context and return the tier-specific model config. This enables image-aware routing where conversations containing screenshots or photos automatically escalate to vision-capable models.

pub fn create_tier_driver( config: &ModelConfig, ) -> PunchResult<Arc<dyn LlmDriver>>

Create an LLM driver for a routed model config.