yera.config.schema.models
Model schema — universe of all discovered models and runtime catalog view.
Symbols
AnthropicLLMCapabilities
LLMCapabilitiesCapabilities specific to Anthropic models.
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
Discriminator for Anthropic capability settings.
Whether the model supports native structured output.
Whether system instructions can appear after the conversation has begun.
AnthropicLLMInference
LLMInferenceAnthropic inference settings model.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
Discriminator for Anthropic inference settings.
How reasoning output is represented in responses.
Default Anthropic reasoning effort.
Whether prompt caching is enabled by default.
Default lifetime of cached prompt content.
AwsLLMCapabilities
LLMCapabilitiesCapabilities of AWS Bedrock LLMs.
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
Discriminator for AWS Bedrock capability settings.
AWS Bedrock API used to invoke the model.
AwsLLMInference
LLMInferenceAWS Bedrock inference settings model.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
Discriminator for AWS Bedrock inference settings.
BaseModelConfig
BaseModelFields common to all model types.
Raises
If id has fewer than three dot-delimited segments.
Fields
Yera-internal dot-delimited identifier. Must have at least three
segments: <provider>.<author>.<model_name>, with optional
further nesting for variants e.g.
llama_cpp.meta.llama3.8b.q4_k_m. Does not encode the
connection name — that is stored separately in connection.
Provider-side model identifier passed directly to the API
or library, e.g. "claude-sonnet-4-20250514" or a HuggingFace
repo id like "meta-llama/Meta-Llama-3-8B-GGUF".
Human-readable display name, e.g. "Claude Sonnet 4".
Key into :class:~yera.config.schema.providers.Providers
identifying which provider serves this model, e.g. "aws".
Name of the provider connection under which this model
was discovered, e.g. "work". Together with provider,
uniquely identifies the runtime interface and credentials to use.
Who created the model, e.g. "anthropic" or "meta".
Distinct from provider — a Meta model can be served via AWS Bedrock.
Methods
BaseModelConfig.id_has_minimum_depth
id_has_minimum_depth(
v: str,
) → strCheck that id has at least three dot-delimited segments.
Parameters
The id string to validate.
Returns
The validated id string.
Raises
If the id has fewer than three segments.
BaseModelConfig.id_prefix_matches_provider
id_prefix_matches_provider() → BaseModelConfigCheck that the first segment of id matches provider.
Returns
The validated model config.
Raises
If the first segment of id does not match provider.
EmbeddingsConfig
BaseModelConfigConfig for an embeddings model.
Fields
Yera-internal dot-delimited identifier. Must have at least three
segments: <provider>.<author>.<model_name>, with optional
further nesting for variants e.g.
llama_cpp.meta.llama3.8b.q4_k_m. Does not encode the
connection name — that is stored separately in connection.
Provider-side model identifier passed directly to the API
or library, e.g. "claude-sonnet-4-20250514" or a HuggingFace
repo id like "meta-llama/Meta-Llama-3-8B-GGUF".
Human-readable display name, e.g. "Claude Sonnet 4".
Key into :class:~yera.config.schema.providers.Providers
identifying which provider serves this model, e.g. "aws".
Name of the provider connection under which this model
was discovered, e.g. "work". Together with provider,
uniquely identifies the runtime interface and credentials to use.
Who created the model, e.g. "anthropic" or "meta".
Distinct from provider — a Meta model can be served via AWS Bedrock.
Discriminator field, always "embedding".
Size of the embedding vector produced by the model.
Maximum number of tokens the model can embed in a single request.
GeminiLLMCapabilities
LLMCapabilitiesCapabilities reported or looked up for a Gemini model.
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
Discriminator for Gemini capability settings.
Whether the model supports native structured output.
Whether the model supports tool calls.
Whether requests can control how the model selects tools.
GeminiLLMInference
LLMInferenceInference settings for Gemini Developer API models.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
Discriminator for Gemini inference settings.
Default number of highest-probability tokens considered.
Whether thought summaries are included by default.
GenericLLMCapabilities
LLMCapabilitiesCapabilities for a generic LLM (no provider-specific features).
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
Discriminator for capabilities without provider-specific behavior.
GenericLLMInference
LLMInferenceDefault inference settings representation.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
Discriminator for inference settings without provider-specific behavior.
LlamaCppLLMCapabilities
LLMCapabilitiesCapabilities of llama-cpp LLMs.
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
Discriminator for llama.cpp capability settings.
Whether the model supports structured output through llama.cpp.
LlamaCppLLMInference
LLMInferenceLlama-cpp inference settings model.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
Discriminator for llama.cpp inference settings.
Number of model layers assigned to the GPU.
Default llama-cpp context length.
LLMCapabilities
BaseModelAnthropicLLMCapabilities, AwsLLMCapabilities, GeminiLLMCapabilities, GenericLLMCapabilities, LlamaCppLLMCapabilities, MistralLLMCapabilities, OpenAILLMCapabilities, OpenRouterLLMCapabilitiesBase for all LLM capabilities models.
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
LLMConfig
BaseModelConfigConfig for a large language model.
Fields
Yera-internal dot-delimited identifier. Must have at least three
segments: <provider>.<author>.<model_name>, with optional
further nesting for variants e.g.
llama_cpp.meta.llama3.8b.q4_k_m. Does not encode the
connection name — that is stored separately in connection.
Provider-side model identifier passed directly to the API
or library, e.g. "claude-sonnet-4-20250514" or a HuggingFace
repo id like "meta-llama/Meta-Llama-3-8B-GGUF".
Human-readable display name, e.g. "Claude Sonnet 4".
Key into :class:~yera.config.schema.providers.Providers
identifying which provider serves this model, e.g. "aws".
Name of the provider connection under which this model
was discovered, e.g. "work". Together with provider,
uniquely identifies the runtime interface and credentials to use.
Who created the model, e.g. "anthropic" or "meta".
Distinct from provider — a Meta model can be served via AWS Bedrock.
Discriminator field, always "llm".
Maximum number of tokens the model can process in a single request (prompt + completion).
Maximum number of tokens the model can generate.
Features and input modes supported by the model.
Default provider-specific settings used when invoking the model.
LLMInference
BaseModelAnthropicLLMInference, AwsLLMInference, GeminiLLMInference, GenericLLMInference, LlamaCppLLMInference, MistralLLMInference, OpenAILLMInference, OpenRouterLLMInferenceBase class for all LLM inference settings models.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
MistralLLMCapabilities
LLMCapabilitiesCapabilities of Mistral LLMs.
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
Discriminator for Mistral capability settings.
MistralLLMInference
LLMInferenceMistral inference settings model.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
Discriminator for Mistral inference settings.
ModelCatalogue
BaseModelRuntime view of the model universe filtered to the active Yera profile.
Never serialized — constructed by :meth:ModelsUniverse.for_profile.
Contains only models whose provider and connection match the
active Yera profile's provider map.
Fields
Catalogue of large language models visible under the active profile.
Catalogue of embeddings models visible under the active profile.
Catalogue of text-to-speech models visible under the active profile.
Catalogue of speech-to-text models visible under the active profile.
Catalogue of reranker models visible under the active profile.
ModelsUniverse
BaseModelAll discovered models across all provider connections.
Serialized to and from yera.toml under [models]. Contains no
defaults — those live in :class:~yera.config.schema.profiles.Profile.
Fields
All discovered large language models.
All discovered embeddings models.
All discovered text-to-speech models.
All discovered speech-to-text models.
All discovered reranker models.
Methods
ModelsUniverse.for_profile
for_profile(
profile: Profile,
) → ModelCatalogueBuild a filtered runtime ModelCatalogue for the active Yera profile.
Filters each model type list to models whose provider and
connection match an entry in the profile's provider map, then
inserts per-type defaults from the profile's model_defaults.
Parameters
The active Yera profile to filter and resolve against.
Returns
A ModelCatalogue containing only models visible under the active profile, with defaults populated from the profile.
OpenAILLMCapabilities
LLMCapabilitiesCapabilities of OpenAI LLMs.
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
Discriminator for OpenAI capability settings.
OpenAILLMInference
LLMInferenceOpenAI inference settings model.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
Discriminator for OpenAI inference settings.
Default level of detail in reasoning summaries.
OpenRouterLLMCapabilities
LLMCapabilitiesCapabilities reported for an OpenRouter model.
Fields
Whether the model accepts image input.
Whether the model accepts PDF document input.
The reasoning control supported by the model.
Reasoning levels accepted by the model.
Whether reasoning must remain enabled for the model.
Discriminator for OpenRouter capability settings.
Whether the model supports native structured output.
Whether the model supports tool calls.
Whether requests can control how the model selects tools.
Whether requests can set an explicit reasoning-token budget.
OpenRouterLLMInference
LLMInferenceInference settings for models invoked through OpenRouter.
Fields
Default maximum number of output tokens.
Default sampling temperature.
Default nucleus-sampling probability.
Default reasoning level when supported by the model.
Default token budget reserved for reasoning.
Discriminator for OpenRouter inference settings.
RerankerConfig
BaseModelConfigConfig for a reranker model.
Fields
Yera-internal dot-delimited identifier. Must have at least three
segments: <provider>.<author>.<model_name>, with optional
further nesting for variants e.g.
llama_cpp.meta.llama3.8b.q4_k_m. Does not encode the
connection name — that is stored separately in connection.
Provider-side model identifier passed directly to the API
or library, e.g. "claude-sonnet-4-20250514" or a HuggingFace
repo id like "meta-llama/Meta-Llama-3-8B-GGUF".
Human-readable display name, e.g. "Claude Sonnet 4".
Key into :class:~yera.config.schema.providers.Providers
identifying which provider serves this model, e.g. "aws".
Name of the provider connection under which this model
was discovered, e.g. "work". Together with provider,
uniquely identifies the runtime interface and credentials to use.
Who created the model, e.g. "anthropic" or "meta".
Distinct from provider — a Meta model can be served via AWS Bedrock.
Discriminator field, always "reranker".
STTConfig
BaseModelConfigConfig for a speech-to-text model.
Fields
Yera-internal dot-delimited identifier. Must have at least three
segments: <provider>.<author>.<model_name>, with optional
further nesting for variants e.g.
llama_cpp.meta.llama3.8b.q4_k_m. Does not encode the
connection name — that is stored separately in connection.
Provider-side model identifier passed directly to the API
or library, e.g. "claude-sonnet-4-20250514" or a HuggingFace
repo id like "meta-llama/Meta-Llama-3-8B-GGUF".
Human-readable display name, e.g. "Claude Sonnet 4".
Key into :class:~yera.config.schema.providers.Providers
identifying which provider serves this model, e.g. "aws".
Name of the provider connection under which this model
was discovered, e.g. "work". Together with provider,
uniquely identifies the runtime interface and credentials to use.
Who created the model, e.g. "anthropic" or "meta".
Distinct from provider — a Meta model can be served via AWS Bedrock.
Discriminator field, always "stt".
TTSConfig
BaseModelConfigConfig for a text-to-speech model.
Fields
Yera-internal dot-delimited identifier. Must have at least three
segments: <provider>.<author>.<model_name>, with optional
further nesting for variants e.g.
llama_cpp.meta.llama3.8b.q4_k_m. Does not encode the
connection name — that is stored separately in connection.
Provider-side model identifier passed directly to the API
or library, e.g. "claude-sonnet-4-20250514" or a HuggingFace
repo id like "meta-llama/Meta-Llama-3-8B-GGUF".
Human-readable display name, e.g. "Claude Sonnet 4".
Key into :class:~yera.config.schema.providers.Providers
identifying which provider serves this model, e.g. "aws".
Name of the provider connection under which this model
was discovered, e.g. "work". Together with provider,
uniquely identifies the runtime interface and credentials to use.
Who created the model, e.g. "anthropic" or "meta".
Distinct from provider — a Meta model can be served via AWS Bedrock.
Discriminator field, always "tts".
TypedCatalogue
BaseModel, Generic[M]A filtered runtime view of models of one type for the active Yera profile.
Never serialized — constructed at runtime by
:meth:ModelsUniverse.for_profile. The default field is populated
from the active :class:~yera.config.schema.profiles.Profile's
model_defaults.
Fields
Models of this type visible under the active Yera profile.
Yera id of the default model for this type, sourced from
the active Yera profile. None if no default is configured —
a runtime error when a default is actually needed.
Methods
TypedCatalogue.default_exists
default_exists() → TypedCatalogueCheck that the default model id exists in models.
Returns
The validated catalogue.
Raises
If default is set but not present in models.