Struct ModelState 

pub struct ModelState {
    pub model: Arc<RwLock<Box<dyn Model>>>,
    pub model_id: String,
    pub model_name: String,
    pub vision_supported: Option<bool>,
    pub supported_reasoning: ReasoningCapability,
    pub base_config: ModelConfig,
}

Model state - LLM configuration and identity

Fields

model: Arc<RwLock<Box<dyn Model>>>

model_id: String

model_name: String

vision_supported: Option<bool>

Vision support state:

• Some(true) = model supports vision
• Some(false) = model does not support vision (detected from error)
• None = unknown (optimistic default)

supported_reasoning: ReasoningCapability

Cached snapshot of the underlying model’s supports_reasoning capability. Lives here so the sync render path doesn’t have to try_read() the tokio::sync::RwLock<Box<dyn Model>> once per frame to compute snap-divergence (Step 5b). Refreshed on /model switch by the slash-command handler.

base_config: ModelConfig

Base model configuration from app config. Used by build_config() to produce API-ready ModelConfig with runtime-only fields set.

Implementations

impl ModelState

pub fn new( model: Box<dyn Model>, model_id: String, base_config: ModelConfig, ) -> Self

pub fn model_ref(&self) -> &Arc<RwLock<Box<dyn Model>>>

Get a reference to the model for reading

pub fn cycle_reasoning(&mut self) -> ReasoningLevel

Cycle the reasoning level for the next chat call. Cycle order: None → Low → Medium → High → Max → None. Minimal and XHigh are omitted from the cycle (specialist tiers restricted to specific providers — OpenAI GPT-5 minimal, GPT-5.2/Opus-4.7 xhigh). Both remain reachable via /reasoning <level>. Returns the new level so the caller can render a status message and persist.

pub fn set_reasoning(&mut self, level: ReasoningLevel)

Set the reasoning level explicitly (used by /reasoning <level>).

pub fn build_config(&self) -> ModelConfig

Build a ModelConfig for API calls using current model state. Clones the base config and sets runtime-only fields.