Struct ModelConfig 

pub struct ModelConfig {
    pub model_path: Option<PathBuf>,
    pub remote_model: Option<String>,
    pub model_repo: Option<String>,
    pub model_quantization: Option<String>,
    pub max_tokens: u32,
    pub temperature: f32,
    pub system_prompt: String,
    pub context_window: Option<usize>,
}

LLM model configuration.

Fields

model_path: Option<PathBuf>

Path to local model file (GGUF/APR/SafeTensors).

remote_model: Option<String>

Remote model identifier (Phase 2, for spillover).

model_repo: Option<String>

HuggingFace repo ID for auto-pull (Phase 2). When set and model_path is None, resolves via apr pull.

model_quantization: Option<String>

Quantization variant for auto-pull (e.g., q4_k_m).

max_tokens: u32

Maximum tokens per completion.

temperature: f32

Sampling temperature.

system_prompt: String

System prompt injected at start of conversation.

context_window: Option<usize>

Context window size override (auto-detected if None).

Implementations

impl ModelConfig

pub fn resolve_model_path(&self) -> Option<PathBuf>

Resolve the effective model path from explicit config only.

Resolution order:

1. Explicit model_path — return as-is
2. model_repo — resolve via pacha cache
3. Neither — return None

Note: auto-discovery from standard paths is done separately in cmd_code (via discover_model()) to avoid side effects in agent manifest validation and tests.

pub fn resolve_model_path_with_discovery(&self) -> Option<PathBuf>

Resolve model path with auto-discovery fallback.

Same as resolve_model_path() but also scans standard paths (~/.apr/models/, ~/.cache/huggingface/, ./models/) for APR/GGUF files. Used by cmd_code for the interactive REPL.

pub fn needs_pull(&self) -> Option<&str>

Check if model needs to be downloaded (auto-pull).

Returns Some(repo) if model_repo is set but the resolved cache path does not exist on disk.

pub fn discover_model() -> Option<PathBuf>

Discover a local model by scanning standard paths.

Search order (per apr-code.md §5.1):

1. ~/.apr/models/
2. ~/.cache/huggingface/ (hub models)
3. ./models/ (project-local)

Within each directory, prefer .apr over .gguf (APR is the stack’s native format — faster loading, row-major layout). Files sorted by modification time (newest first).

PMAT-150 (Jidoka): APR files are validated at discovery time — if an APR file lacks an embedded tokenizer, it is deprioritized so GGUF files are tried first. This prevents the user from hitting a dead-end error when the only APR model is broken.

pub fn model_search_dirs() -> Vec<PathBuf>

Standard model search directories.

pub fn auto_pull(&self, timeout_secs: u64) -> Result<PathBuf, AutoPullError>

Auto-pull model via apr pull subprocess.

Invokes apr pull <repo> with a configurable timeout. The apr CLI handles caching internally at ~/.cache/pacha/models/. Returns the resolved cache path on success.

Jidoka: stops on subprocess failure rather than continuing with a missing model.

Trait Implementations

impl Clone for ModelConfig

fn clone(&self) -> ModelConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ModelConfig

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

Formats the value using the given formatter.

impl Default for ModelConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for ModelConfig

where ModelConfig: Default,

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ModelConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.