Skip to content

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

test(inference): add tests for model resolution and profile authority

Steven Enamakel85df758f1f55
diff --git a/crates/openhuman-core/src/inference/provider/factory_crate_native_tests.rs b/crates/openhuman-core/src/inference/provider/factory_crate_native_tests.rs
index fe5b0654d5..f327c2454a 100644
--- a/crates/openhuman-core/src/inference/provider/factory_crate_native_tests.rs
+++ b/crates/openhuman-core/src/inference/provider/factory_crate_native_tests.rs
@@ -209,6 +209,38 @@ async fn one_shot_chat_models_preserve_factory_temperature_as_request_default()
);
}
+#[tokio::test]
+async fn one_shot_cloud_sentinel_resolves_the_configured_byok_provider() {
+ let _guard = crate::inference::inference_test_guard_async().await;
+ let _env = crate::config::TEST_ENV_LOCK.lock().await;
+ let tmp = tempfile::tempdir().expect("isolated workspace");
+ let mut config = config_with_providers_in_tempdir(
+ &tmp,
+ vec![openai_entry("p_openai", "openai")],
+ );
+ config.chat_provider = Some("openai:gpt-4o-mini".into());
+ store_byo_key(&config, "openai", "sk-fixture");
+ let library = crate::core::runtime::CoreContext::for_test_with_host_kind(
+ crate::core::types::HostKind::Library,
+ crate::core::runtime::DomainSet::full(),
+ Some(tmp.path().join("library")),
+ );
+
+ crate::core::runtime::CoreContext::scope(library, async {
+ for provider in ["cloud", ""] {
+ let (_, model) = create_chat_model_from_string_with_model_id(
+ "chat",
+ provider,
+ &config,
+ 0.2,
+ )
+ .expect("sentinel resolves to the configured BYOK route");
+ assert_eq!(model, "gpt-4o-mini");
+ }
+ })
+ .await;
+}
+
// ── Motion B (#4727): managed-backend crate-native routing ──────────────────
// `create_chat_model` must route the managed OpenHuman backend through the
// crate-native `TinyHumansModel`, whose profile advertises the capabilities
diff --git a/crates/openhuman-core/src/inference/turn_models_tests.rs b/crates/openhuman-core/src/inference/turn_models_tests.rs
index abc6f116c3..73f0a0aa36 100644
--- a/crates/openhuman-core/src/inference/turn_models_tests.rs
+++ b/crates/openhuman-core/src/inference/turn_models_tests.rs
@@ -287,6 +287,85 @@ async fn explicit_provider_string_turn_source_uses_scoped_managed_model() {
.await;
}
+#[tokio::test]
+async fn injected_turn_profile_is_authoritative_and_skips_unselected_routes() {
+ use crate::agent::host_overrides::HostOverrides;
+ use crate::core::runtime::{ContextOverlay, CoreContext, DispatchCtx, DomainSet};
+ use crate::core::slots::Extensions;
+ use crate::core::types::HostKind;
+ use tinyinference_llm::model::ModelProfile;
+ use std::sync::atomic::{AtomicUsize, Ordering};
+
+ struct ProfiledModel {
+ profile: ModelProfile,
+ }
+ #[async_trait]
+ impl ChatModel<()> for ProfiledModel {
+ fn profile(&self) -> Option<&ModelProfile> {
+ Some(&self.profile)
+ }
+ async fn invoke(
+ &self,
+ _state: &(),
+ _request: ModelRequest,
+ ) -> tinyinference_llm::Result<ModelResponse> {
+ Ok(ModelResponse::assistant("injected"))
+ }
+ }
+
+ let _guard = crate::inference::inference_test_guard_async().await;
+ let builds = Arc::new(AtomicUsize::new(0));
+ let count = Arc::clone(&builds);
+ let extensions = Extensions::new();
+ crate::inference::registry::register_chat_model_in(
+ &extensions,
+ crate::inference::provider::factory::PROVIDER_OPENHUMAN,
+ crate::inference::registry::ChatModelEntry {
+ build: Arc::new(move |_| {
+ count.fetch_add(1, Ordering::SeqCst);
+ Ok((Arc::new(NamedModel("unselected")), "unselected".into()))
+ }),
+ probe: None,
+ models: None,
+ },
+ );
+ let root = CoreContext::for_test_with_host_kind_and_extensions(
+ HostKind::Library,
+ DomainSet::full(),
+ None,
+ extensions,
+ );
+ let mut profile = ModelProfile::default();
+ profile.provider = Some("ollama:injected".into());
+ profile.tool_calling = true;
+ profile.modalities.image_in = true;
+ profile.max_input_tokens = Some(123_456);
+ let mut overrides = HostOverrides::default();
+ overrides.model = Some(Arc::new(ProfiledModel { profile }));
+ let context = root.derive_with(
+ ContextOverlay::new(
+ crate::config::Config {
+ chat_provider: Some("openhuman".into()),
+ ..Default::default()
+ },
+ DomainSet::full(),
+ Default::default(),
+ )
+ .host_overrides(Arc::new(overrides)),
+ );
+ let ctx = DispatchCtx::new(context);
+ let models = TurnModelSource::new_crate_native(&ctx, "chat", Arc::new(Default::default()))
+ .build("hint:chat", 0.2, None, None)
+ .expect("host injected model builds");
+
+ assert_eq!(models.provider_id(), "ollama:injected");
+ assert!(models.native_tools());
+ assert!(models.supports_vision());
+ assert!(models.is_local());
+ assert_eq!(models.context_window(), Some(123_456));
+ assert_eq!(builds.load(Ordering::SeqCst), 0);
+}
+
#[tokio::test]
async fn explicit_turn_context_without_provider_does_not_borrow_ambient_provider() {
use crate::core::runtime::{CoreContext, DispatchCtx, DomainSet};