Skip to content

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

fix(inference): stop inspection factory from reading caller extensions

Steven Enamakel9b5135321914
diff --git a/crates/openhuman-core/src/inference/provider/factory/inspection.rs b/crates/openhuman-core/src/inference/provider/factory/inspection.rs
index ddf50157ff..a634bc35eb 100644
--- a/crates/openhuman-core/src/inference/provider/factory/inspection.rs
+++ b/crates/openhuman-core/src/inference/provider/factory/inspection.rs
@@ -27,9 +27,7 @@ pub(crate) fn create_chat_model_with_identity_in_for_inspection(
temperature: f64,
) -> anyhow::Result<PreparedInspectionChatModel> {
let resolved_context = ctx.resolved();
- let extensions = resolved_context
- .as_deref()
- .map(crate::core::runtime::CoreContext::extensions);
+ let extensions = None;
let mut resolved = provider_for_role(role, config);
if resolved.trim().is_empty() || resolved.trim() == "cloud" {
resolved = resolve_primary_cloud_provider_string(config);
diff --git a/crates/openhuman-core/src/inference/registry_tests.rs b/crates/openhuman-core/src/inference/registry_tests.rs
index 56750fa6b0..0ef60637da 100644
--- a/crates/openhuman-core/src/inference/registry_tests.rs
+++ b/crates/openhuman-core/src/inference/registry_tests.rs
@@ -302,15 +302,20 @@ fn inspection_model_factory_uses_the_callers_context_before_process_fallback() {
Arc::new(move |request| {
Ok(InspectionChatModelBuild {
model: Arc::new(FakeModel),
- model_id: request.context.config.default_model.clone().unwrap(),
+ model_id: format!(
+ "{endpoint}:{}",
+ request.context.config.default_model.as_deref().unwrap()
+ ),
endpoint: format!("https://{endpoint}.example/v1"),
credential: format!("{endpoint}-credential"),
})
})
};
register_inspection_chat_model(PROVIDER_OPENHUMAN, factory("process"));
+ register_chat_model(PROVIDER_OPENHUMAN, fake_entry(Arc::default()));
let a = Extensions::new();
register_inspection_chat_model_in(&a, PROVIDER_OPENHUMAN, factory("context-a"));
+ register_chat_model_in(&a, PROVIDER_OPENHUMAN, fake_entry(Arc::default()));
let context_a = crate::core::runtime::DispatchCtx::new(
crate::core::runtime::CoreContext::for_test_with_host_kind_and_extensions(
crate::core::types::HostKind::Library,
@@ -321,6 +326,7 @@ fn inspection_model_factory_uses_the_callers_context_before_process_fallback() {
);
let b = Extensions::new();
register_inspection_chat_model_in(&b, PROVIDER_OPENHUMAN, factory("context-b"));
+ register_chat_model_in(&b, PROVIDER_OPENHUMAN, fake_entry(Arc::default()));
let context_b = crate::core::runtime::DispatchCtx::new(
crate::core::runtime::CoreContext::for_test_with_host_kind_and_extensions(
crate::core::types::HostKind::Library,
@@ -350,8 +356,9 @@ fn inspection_model_factory_uses_the_callers_context_before_process_fallback() {
)
.expect("process inspection provider");
unregister_inspection_chat_model(PROVIDER_OPENHUMAN);
+ unregister_chat_model(PROVIDER_OPENHUMAN);
- assert_eq!(build_a.model_id, "context-a");
- assert_eq!(build_b.model_id, "context-b");
- assert_eq!(build_process.model_id, "process");
+ assert_eq!(build_a.model_id, "context-a:context-a");
+ assert_eq!(build_b.model_id, "context-b:context-b");
+ assert_eq!(build_process.model_id, "process:process");
}