Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

refactor(inference): resolve provider hooks from ambient context

Steven Enamakel8f5efb840d54
diff --git a/crates/openhuman-core/src/inference/provider_model_tests.rs b/crates/openhuman-core/src/inference/provider_model_tests.rs
index 27ae32254f..82870fb9d8 100644
--- a/crates/openhuman-core/src/inference/provider_model_tests.rs
+++ b/crates/openhuman-core/src/inference/provider_model_tests.rs
@@ -290,11 +290,6 @@ fn native_provider_metadata_resolves_by_context_and_preserves_request_fields() {
provider,
&format!("{label}/alias")
));
- assert!(crate::inference::registry::provider_model_is_raw_for(
- extensions,
- provider,
- &format!("{label}/raw")
- ));
assert!(
crate::inference::registry::provider_model_supports_vision_for(
extensions,
@@ -324,26 +319,42 @@ fn native_provider_metadata_resolves_by_context_and_preserves_request_fields() {
),
Some(window)
);
- assert_eq!(
- crate::inference::registry::provider_default_pricing_for(extensions, provider),
- Some((format!("{label}/priced"), 1.25, 2.5))
- );
let mut request = tinyinference_llm::model::ModelRequest::new(vec![]);
request.model = Some(format!("{label}/raw"));
request.max_tokens = Some(37);
request.metadata = json!({"caller": "kept"});
- let request = crate::inference::registry::provider_request_without_reasoning_for(
- extensions, provider, request,
- );
- assert_eq!(
- request.model.as_deref(),
- Some(format!("{label}/raw").as_str())
- );
- assert_eq!(request.max_tokens, Some(37));
- assert_eq!(
- request.metadata,
- json!({"caller": "kept", "reasoning_removed_by": label})
+ let ambient_extensions = extensions.cloned().unwrap_or_else(Extensions::new);
+ let ambient = crate::core::runtime::CoreContext::for_test_with_extensions(
+ crate::core::runtime::DomainSet::full(),
+ None,
+ ambient_extensions,
);
+ crate::core::runtime::CoreContext::sync_scope(ambient, || {
+ assert!(crate::inference::registry::provider_model_is_raw(
+ provider,
+ &format!("{label}/raw")
+ ));
+ assert_eq!(
+ crate::inference::registry::provider_default_pricing(provider),
+ Some((format!("{label}/priced"), 1.25, 2.5))
+ );
+ let mut request = tinyinference_llm::model::ModelRequest::new(vec![]);
+ request.model = Some(format!("{label}/raw"));
+ request.max_tokens = Some(37);
+ request.metadata = json!({"caller": "kept"});
+ let request = crate::inference::registry::provider_request_without_reasoning(
+ provider, request,
+ );
+ assert_eq!(
+ request.model.as_deref(),
+ Some(format!("{label}/raw").as_str())
+ );
+ assert_eq!(request.max_tokens, Some(37));
+ assert_eq!(
+ request.metadata,
+ json!({"caller": "kept", "reasoning_removed_by": label})
+ );
+ });
};
verify(None, "process", 101);
diff --git a/crates/openhuman-core/src/inference/registry.rs b/crates/openhuman-core/src/inference/registry.rs
index af3779a4d2..8f0ea0c5ea 100644
--- a/crates/openhuman-core/src/inference/registry.rs
+++ b/crates/openhuman-core/src/inference/registry.rs
@@ -101,7 +101,7 @@ impl ProviderContext {
}
}
-pub use tinyinference_llm::providers::overrides::{
+use tinyinference_llm::providers::overrides::{
BoxFuture, CatalogEntry, CatalogLimits, CatalogRequest, ChatModelEntry, ChatModelFactory,
ChatModelRequest, EmbeddingFactory, EmbeddingRequest, ProviderModelMetadata, ProviderOverrides,
ReadinessProbe,
@@ -260,14 +260,6 @@ pub fn provider_default_pricing(provider_id: &str) -> Option<(String, f64, f64)>
(hooks.default_pricing)()
}
-pub(crate) fn provider_default_pricing_for(
- extensions: Option<&Extensions>,
- provider_id: &str,
-) -> Option<(String, f64, f64)> {
- let hooks = chat_model_for(extensions, provider_id)?.models?;
- (hooks.default_pricing)()
-}
-
/// The workload role assigned to a provider-specific model alias.
pub fn provider_model_role(provider_id: &str, model: &str) -> Option<String> {
let hooks = provider_model_hooks(provider_id)?;
@@ -306,19 +298,6 @@ pub fn provider_model_is_raw(provider_id: &str, model: &str) -> bool {
&& provider_model_hooks(provider_id).is_none_or(|hooks| (hooks.is_raw_model)(model))
}
-pub(crate) fn provider_model_is_raw_for(
- extensions: Option<&Extensions>,
- provider_id: &str,
- model: &str,
-) -> bool {
- let model = model.trim();
- !model.is_empty()
- && !model.starts_with("hint:")
- && chat_model_for(extensions, provider_id)
- .and_then(|entry| entry.models)
- .is_none_or(|hooks| (hooks.is_raw_model)(model))
-}
-
/// Whether the provider says `model` supports image input.
pub fn provider_model_supports_vision(provider_id: &str, model: &str) -> bool {
provider_model_hooks(provider_id).is_some_and(|hooks| (hooks.supports_vision)(model))
@@ -390,17 +369,6 @@ pub fn provider_request_without_reasoning(
}
}
-pub(crate) fn provider_request_without_reasoning_for(
- extensions: Option<&Extensions>,
- provider_id: &str,
- request: tinyinference_llm::model::ModelRequest,
-) -> tinyinference_llm::model::ModelRequest {
- match chat_model_for(extensions, provider_id).and_then(|entry| entry.models) {
- Some(hooks) => (hooks.without_reasoning)(request),
- None => request,
- }
-}
-
/// Registers a generic config normalizer, such as a provider-owned migration
/// that must also run after the config schema reaches its current version.
pub type ConfigNormalizer = Arc<dyn Fn(&mut Config) -> bool + Send + Sync>;
diff --git a/crates/openhuman-embed/src/host_internals.rs b/crates/openhuman-embed/src/host_internals.rs
index 564e8c3e5c..99d6585870 100644
--- a/crates/openhuman-embed/src/host_internals.rs
+++ b/crates/openhuman-embed/src/host_internals.rs
@@ -376,10 +376,12 @@ pub mod inference_registry {
register_config_normalizer, register_config_normalizer_in, register_embedding,
register_embedding_in, resolve_provider_model, unregister_backend_error_classifier,
unregister_catalog, unregister_chat_model, unregister_embedding,
- BackendErrorClassification, BackendErrorClassifier, BackendErrorKind, BoxFuture,
- CatalogCallContext, CatalogEntry, CatalogHostRequest, CatalogLimits, CatalogRequest,
- ChatModelEntry, ChatModelFactory, ChatModelRequest, ConfigNormalizer, EmbeddingCallContext,
- EmbeddingFactory, EmbeddingRequest, ProviderContext, ProviderModelMetadata,
+ BackendErrorClassification, BackendErrorClassifier, BackendErrorKind, CatalogCallContext,
+ CatalogHostRequest, ConfigNormalizer, EmbeddingCallContext, ProviderContext,
+ };
+ pub use tinyinference_llm::providers::overrides::{
+ BoxFuture, CatalogEntry, CatalogLimits, CatalogRequest, ChatModelEntry, ChatModelFactory,
+ ChatModelRequest, EmbeddingFactory, EmbeddingRequest, ProviderModelMetadata,
ProviderOverrides, ReadinessProbe,
};
}