Struct ModelRegistry 

pub struct ModelRegistry { /* private fields */ }

All known models with their configuration.

Implementations

impl ModelRegistry

pub fn load(agent_dir: &Path) -> ModelRegistry

Loads the catalog with models.json and auth.json from agent_dir. It ignores a broken models.json and reports it through ModelRegistry::error.

pub fn enable_llama(&mut self)

Adds pi’s llama.cpp provider, as its built-in extension does.

pub fn llama_enabled(&self) -> bool

Whether the llama.cpp provider is present.

pub async fn llama_server(&self) -> Option<(String, Option<String>)>

The llama.cpp server and its key, when the provider is configured.

pub async fn catalog_targets(&self) -> Vec<Target>

What a catalog refresh needs: each refreshable provider, whether it is configured, and its token for gateways that require one.

pub fn models_store(&self) -> Option<&ModelsStore>

Where refreshed catalogs persist, when the registry was loaded from an agent directory.

pub fn apply_catalogs(&mut self, models: IndexMap<String, ProviderModels>)

Replaces the refreshed catalogs of the providers in models.

pub async fn provider_token(&self, provider: &str) -> Option<String>

A token for provider-level calls such as catalog refreshes: the API key or bearer token a request would send, refreshed first when it is an expiring OAuth token.

pub async fn provider_level_auth(&self, provider: &str) -> Option<Auth>

Credentials for provider as a whole, with its models.json headers but no model’s, as pi’s getAuth(providerId). None for a provider the registry does not know.

pub fn builtin() -> ModelRegistry

The catalog alone, with no files.

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

Why models.json could not be applied, if it could not.

pub fn models(&self) -> &[Model]

Every chat model.

pub fn find(&self, provider: &str, id: &str) -> Option<&Model>

A model by provider and id.

pub fn classifiers(&self) -> &[ClassifierModel]

Every classifier model, built-in and refreshed.

pub fn image_models(&self) -> &[ImageModel]

Every image-generation model, built-in and refreshed.

pub fn available_classifiers(&self) -> Vec<&ClassifierModel>

Classifier models whose providers are configured and keep them.

pub fn available_image_models(&self) -> Vec<&ImageModel>

Image models whose providers are configured and keep them.

pub async fn classify( &self, model: &ClassifierModel, context: &ClassifierContext, options: ClassifyOptions, ) -> ClassifierResult

Classifies context with model, resolving the provider’s credentials (pi’s Models.classify). Failures are in the result.

pub async fn generate_images( &self, model: &ImageModel, context: &ImagesContext, options: ImagesOptions, ) -> AssistantImages

Generates images with model, resolving the provider’s credentials (pi’s Models.generateImages). Failures are in the result.

pub fn register_config(&mut self, provider: &str, config: ProviderConfig)

Adds a provider an extension configures, as a models.json entry that replaces any of the same name (pi’s registerProvider).

pub fn is_extension_provider(&self, provider: &str) -> bool

Whether an extension registered provider and has not unregistered it.

pub fn unregister_provider(&mut self, provider: &str)

Removes the provider an extension registered as provider, with its sign-ins, availability and implementations, and restores the built-in or models.json provider it replaced (pi’s unregisterProvider). Does nothing for a provider no extension registered.

pub fn replace_models( &mut self, provider: &str, models: Vec<ModelDefinition>, ) -> Result<(), String>

Replaces the models an extension provider configures with ones its refreshModels returned. Fails, changing nothing, when a model lacks an API or base URL, as pi validates a refreshed list before using it.

pub fn register_operations( &mut self, provider: &str, apis: impl IntoIterator<Item = String>, implementation: Arc<dyn ModelOperations>, )

Uses implementation for provider’s classifier and image models of apis (pi’s classifiers and images of an extension’s provider).

pub fn register_key_auth(&mut self, provider: &str, auth: Arc<dyn KeyAuth>)

Resolves provider’s API key with auth instead of the environment, and signs in with its login. The provider is not configured without a stored credential until ModelRegistry::set_availability finds a key.

pub fn key_auth(&self, provider: &str) -> Option<&Arc<dyn KeyAuth>>

provider’s API-key auth from an extension, if it has one.

pub fn set_availability(&mut self, provider: &str, availability: Availability)

Applies what extension provider provider makes available for the stored credential: where its key auth finds a key, and the models its filters keep.

pub fn modify_models(&mut self, provider: &str, models: Option<Vec<Model>>)

Uses models as provider’s chat models, as an extension’s legacy oauth.modifyModels made them for its signed-in account. None restores the provider’s own.

pub async fn refresh_credential(&self, provider: &str) -> Option<Credential>

The credential a catalog refresh of provider passes to the network, as pi’s resolveRefreshCredential: a stored OAuth credential, refreshed when it has expired, else the API key a request would use.

pub fn register_provider(&mut self, provider: &str, models: Vec<Model>)

Adds models from an extension provider, replacing the provider’s models.

pub fn register_oauth(&mut self, provider: &str, flow: Arc<dyn OAuthProvider>)

Uses flow for provider’s sign-in and refresh, replacing the built-in.

pub fn oauth_flow(&self, provider: &str) -> Option<Arc<dyn OAuthProvider>>

The sign-in for provider: a registered one, a models.json Radius gateway’s, else yapi’s built-in.

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

The gateway of a models.json provider with "oauth": "radius": its base URL without a trailing /v1.

pub fn set_runtime_key(&mut self, provider: &str, key: String)

Uses key for provider for this process, ahead of every stored credential.

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

provider’s credential as pi’s RuntimeCredentials.read gives it to providers: a runtime key as an API key credential, else the stored one.

pub fn store(&self) -> &Arc<CredentialStore>

The credential store behind this registry.

pub fn has_auth(&self, provider: &str) -> bool

Whether some credential is configured for provider, without running commands.

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

pi’s failure for provider while auth.json cannot be read: requests that would consult it fail instead of using other credentials.

pub fn uses_command_key(&self, provider: &str) -> bool

Whether provider’s credential comes from a !command, whose output decides whether there is a key at all.

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

How provider is configured, as pi’s getProviderAuthStatus labels it in /login: runtime, stored, models_json_command, models_json_key, fallback (an extension’s key), or the environment variables the key comes from. None when it is not configured.

pub fn is_using_oauth(&self, provider: &str) -> bool

Whether provider authenticates with a stored OAuth credential. A runtime key takes its place.

pub fn stored_credentials(&self) -> Vec<(String, CredentialKind)>

Stored credentials by provider, in file order.

pub fn available(&self) -> Vec<&Model>

Models with credentials, as pi’s getAvailable. A GitHub Copilot sign-in limits its models to those the account enables, and an extension provider’s filterModels to those it keeps.

pub fn available_of_all_types(&self) -> Vec<&Model>

Chat models with credentials as pi’s getAvailableOfType("chat") lists them, where a provider’s filterAllModels decides over its filterModels.

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

The API key a request to provider would use, when it is known without running a command or signing in: a runtime key, a stored or configured template value, or the environment.

pub async fn auth(&self, model: &Model) -> Auth

Credentials for a request to model, with its configured headers. A stored OAuth token about to expire is refreshed first.

pub async fn auth_valid_for( &self, model: &Model, min_validity_ms: Option<u64>, ) -> Auth

ModelRegistry::auth with pi’s minOAuthValidityMs: a stored OAuth token valid for less than five minutes or min_validity_ms is refreshed first, and with min_validity_ms a refreshed token valid for less than that is an error.

pub fn provider_name(&self, provider: &str) -> String

The display name of a provider: its models.json or extension name, else the built-in name, else the name of its registered sign-in, else its id.

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

The models.json or extension configuration of provider.

pub fn registered_by_extension(&self, provider: &str) -> bool

Whether an extension registered provider.

pub fn auth_header(&self, provider: &str) -> bool

Whether models.json or an extension sets authHeader for provider.

pub fn configures_api_key(&self, provider: &str) -> bool

Whether models.json or an extension configures an apiKey for provider, even an empty one.

pub async fn login( &self, provider: &str, kind: LoginKind, interaction: &Interaction, options: &LoginOptions, ) -> Result<Credential, AuthError>

Runs a sign-in and stores the credential. An API key login asks for the key, with an extension provider’s own login when it has one, and an OAuth login runs the provider’s flow.

pub async fn logout(&self, provider: &str) -> Result<(), AuthError>

Removes provider’s stored credential.

pub fn auth_path(&self) -> Option<&Path>

Where auth.json lives.

Trait Implementations

impl Clone for ModelRegistry

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ModelRegistry

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

Formats the value using the given formatter.

impl Default for ModelRegistry

fn default() -> Self

Returns the “default value” for a type.