Skip to content

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

feat(inference): implement tracking and admission gate methods

Steven Enamakel6f90d114c3dc
diff --git a/crates/openhuman-core/src/agent/session_host/driver/grounded_close.rs b/crates/openhuman-core/src/agent/session_host/driver/grounded_close.rs
index ef2f7ac37f..a0de5349ab 100644
--- a/crates/openhuman-core/src/agent/session_host/driver/grounded_close.rs
+++ b/crates/openhuman-core/src/agent/session_host/driver/grounded_close.rs
@@ -170,6 +170,12 @@ impl Drop for SupplementalCall {
#[async_trait::async_trait]
impl ChatModel<()> for SummarizerSourceModel<'_> {
+ fn tracks_physical_attempts(&self) -> bool {
+ // The selected provider is resolved afresh for each call and can be a
+ // host-injected model. Do not claim hook coverage for every route.
+ false
+ }
+
async fn invoke(
&self,
state: &(),
diff --git a/crates/openhuman-core/src/inference/provider/factory/chat_model.rs b/crates/openhuman-core/src/inference/provider/factory/chat_model.rs
index 4b067205b4..269f6c2fbb 100644
--- a/crates/openhuman-core/src/inference/provider/factory/chat_model.rs
+++ b/crates/openhuman-core/src/inference/provider/factory/chat_model.rs
@@ -316,6 +316,14 @@ impl ChatModel<()> for DefaultTemperatureChatModel {
self.inner.cache_identity()
}
+ fn tracks_physical_attempts(&self) -> bool {
+ self.inner.tracks_physical_attempts()
+ }
+
+ fn has_pre_invocation_admission_gate(&self) -> bool {
+ self.inner.has_pre_invocation_admission_gate()
+ }
+
async fn invoke(
&self,
state: &(),
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 466481c27c..29ab6af4fe 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
@@ -253,6 +253,10 @@ fn default_temperature_wrapper_preserves_physical_attempt_tracking() {
true
}
+ fn has_pre_invocation_admission_gate(&self) -> bool {
+ true
+ }
+
async fn invoke(
&self,
_state: &(),
@@ -273,6 +277,7 @@ fn default_temperature_wrapper_preserves_physical_attempt_tracking() {
.expect("test override model");
assert!(wrapped.tracks_physical_attempts());
+ assert!(wrapped.has_pre_invocation_admission_gate());
}
// ── Motion B (#4727): managed-backend crate-native routing ──────────────────
diff --git a/crates/openhuman-tinyhumans/src/inference/managed_calls.rs b/crates/openhuman-tinyhumans/src/inference/managed_calls.rs
index f16c9db35b..54c88a970f 100644
--- a/crates/openhuman-tinyhumans/src/inference/managed_calls.rs
+++ b/crates/openhuman-tinyhumans/src/inference/managed_calls.rs
@@ -64,6 +64,12 @@ impl ChatModel<()> for OpenHumanBackendModel {
Some(&self.profile)
}
+ fn tracks_physical_attempts(&self) -> bool {
+ // Every request is forwarded to the native OpenAI-compatible adapter,
+ // which checks the physical-attempt hook at its HTTP send boundary.
+ true
+ }
+
fn supports_input(
&self,
modality: tinyinference_llm::model::InputModality,