Struct ModelRegistry 

pub struct ModelRegistry { /* private fields */ }

Indexed view over a ModelConfiguration with identifier-normalised lookup.

ModelRegistry owns the merged catalog and two auxiliary indices — by API identifier and by provider — populated at construction time. Construct one with ModelRegistry::load, which performs the layered YAML load described on ModelConfiguration. Most callers use the process-wide singleton returned by get_model_registry rather than loading their own instance.

Implementations

impl ModelRegistry

pub fn load() -> Result<Self>

Loads the model registry, layering an optional user-side catalog over the embedded one.

Lookup order (highest precedence wins):

1. OMNI_DEV_MODELS_YAML — explicit override path; short-circuits 2 & 3.
2. ./.omni-dev/models.yaml — project-local catalog (if present).
3. ~/.omni-dev/models.yaml — user-level catalog (if present).
4. Embedded src/templates/models.yaml — always present, lowest layer.

Missing user-side files fall through silently. Malformed user-side files log an error and are skipped. A malformed embedded catalog is a hard failure (compile-time invariant).

pub fn load_layered_from_paths( project_path: Option<&Path>, user_path: Option<&Path>, override_path: Option<&Path>, ) -> Result<Self>

Loads the registry with explicit paths for the user-side layers.

Exposed primarily for testing — the public entry point is load().

pub fn config(&self) -> &ModelConfiguration

Returns the merged model configuration.

pub fn get_model_spec(&self, api_identifier: &str) -> Option<&ModelSpec>

Returns the model specification for the given API identifier.

pub fn get_max_output_tokens(&self, api_identifier: &str) -> usize

Returns the max output tokens for a model, with fallback to provider defaults.

pub fn get_input_context(&self, api_identifier: &str) -> usize

Returns the input context limit for a model, with fallback to provider defaults.

pub fn supports_structured_output(&self, api_identifier: &str) -> bool

Returns whether the model supports structured JSON-schema output (output_config.format).

Backs the schema-capability gate for the direct Anthropic and Bedrock backends (#1119). Resolves through the same identifier normalization as the other lookups, so Bedrock/region-prefixed forms map to the same ModelSpec. Any model not present in the catalog — or present but not flagged — returns false, keeping it on the YAML path rather than risking a 400 from an unsupported output_config.

pub fn is_legacy_model(&self, api_identifier: &str) -> bool

Checks if a model is legacy.

pub fn get_all_models(&self) -> &[ModelSpec]

Returns all available models.

pub fn get_models_by_provider(&self, provider: &str) -> Vec<&ModelSpec>

Returns models filtered by provider.

pub fn is_known_model(&self, identifier: &str) -> bool

Reports whether identifier names a model this registry knows.

Goes through ModelRegistry::get_model_spec, so Bedrock-style identifiers (us.anthropic.…-v1:0) are normalised before lookup and user-supplied models.yaml entries count as known.

pub fn known_identifiers(&self, provider: &str) -> Vec<&str>

Returns the non-legacy API identifiers for a provider, sorted.

Intended for error messages that need to show the user what they could have typed. Legacy models are omitted: they still resolve, but should not be suggested.

pub fn get_models_by_provider_and_tier( &self, provider: &str, tier: &str, ) -> Vec<&ModelSpec>

Returns models filtered by provider and tier.

pub fn get_default_model(&self, provider: &str) -> Option<&str>

Returns the default model identifier for a provider, as defined in models.yaml.

pub fn get_provider_config(&self, provider: &str) -> Option<&ProviderConfig>

Returns the provider configuration.

pub fn get_tier_info(&self, provider: &str, tier: &str) -> Option<&TierInfo>

Returns tier information for a provider.

pub fn get_beta_headers(&self, api_identifier: &str) -> &[BetaHeader]

Returns the beta headers for a model.

pub fn get_max_output_tokens_with_beta( &self, api_identifier: &str, beta_value: &str, ) -> usize

Returns the max output tokens for a model with a specific beta header active.

pub fn get_input_context_with_beta( &self, api_identifier: &str, beta_value: &str, ) -> usize

Returns the input context for a model with a specific beta header active.

pub fn get_model_family(&self, api_identifier: &str) -> ModelFamily

Classifies a raw model identifier into its ModelFamily.

Resolves through Self::get_model_spec first, so Bedrock/region- prefixed and versioned identifiers normalize the same way every other lookup on this registry does; the family word (Fable/Opus/Sonnet/ Haiku) is then matched, case-insensitively, against the resolved spec’s human-readable ModelSpec::model name. A registry miss falls back to matching directly against the raw identifier, so a brand-new model not yet in the catalog still classifies correctly as long as its id contains the family word. No match on either ⇒ ModelFamily::Unknown.