Struct RunningTarget 

pub struct RunningTarget {
    pub base_url: String,
    pub port: u16,
    pub model_id: u32,
    pub model_name: String,
    pub effective_ctx: u64,
    pub just_started: bool,
    pub slot_restore_supported: bool,
    pub cache_ram_health: CacheRamHealth,
    pub narration: Option<LaunchNarration>,
}

Target information for a running model instance.

This struct contains all information needed to route requests to a running llama-server instance.

Fields

base_url: String

Full URL to the server (e.g., http://127.0.0.1:5500). Future-proof for non-localhost deployments.

port: u16

Port the server is listening on.

model_id: u32

Database ID of the model.

model_name: String

Human-readable model name (for logging/headers).

effective_ctx: u64

Actual context size being used.

just_started: bool

True when this instance was freshly spawned (restart or cold start).

slot_restore_supported: bool

Whether llama-server’s disk slot save/restore can actually resume this model, i.e. its KV memory retains the full token history.

False for sliding-window, hybrid, and recurrent architectures (see crate::domain::kv_memory_is_partial): the slot file carries KV state and tokens but not the server’s context checkpoints, so a restore leaves the slot unable to resume and llama-server re-prefills the whole prompt. Callers skip the disk slot layer when this is false and let the in-RAM prompt cache — which does keep checkpoints — handle conversation switching.

cache_ram_health: CacheRamHealth

How healthy the host-RAM prompt cache budget (--cache-ram) resolved for this launch is.

Classified once at spawn (where the budget arithmetic and the auto-vs-explicit distinction are both in scope) and carried here so user-facing surfaces can report it without re-deriving thresholds. See crate::domain::classify_cache_ram.

narration: Option<LaunchNarration>

What this launch decided, and why (see crate::domain::LaunchNarration).

Carried on the target for the same reason as Self::cache_ram_health: the resolutions and their provenance exist only at spawn, so anything downstream that wants to explain the running model has no way to recover them otherwise. None for targets that did not come from a gglib launch.

Implementations

impl RunningTarget

pub fn local( port: u16, model_id: u32, model_name: String, effective_ctx: u64, just_started: bool, ) -> Self

Create a new RunningTarget for a local server.

slot_restore_supported defaults to true (the full-attention case) and cache_ram_health to CacheRamHealth::LlamaDefault (no flag emitted); callers that know the launch’s actual resolution narrow them with Self::with_slot_restore_supported and Self::with_cache_ram_health.

pub fn with_narration(self, narration: LaunchNarration) -> Self

Attach the narration of the launch that produced this target.

pub const fn with_slot_restore_supported(self, supported: bool) -> Self

Set whether disk slot restore can resume this model.

pub const fn with_cache_ram_health(self, health: CacheRamHealth) -> Self

Set the resolved host-RAM prompt cache health for this launch.

Trait Implementations

impl Clone for RunningTarget

fn clone(&self) -> RunningTarget

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for RunningTarget

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.