Skip to content

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

refactor(inference): pass provider context to model resolution and probes

Steven Enamakelbd8de1bf6376
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 6040e83047..eb8754ee2a 100644
--- a/crates/openhuman-core/src/inference/provider/factory/chat_model.rs
+++ b/crates/openhuman-core/src/inference/provider/factory/chat_model.rs
@@ -505,10 +505,14 @@ pub async fn probe_inference_readiness(
log::debug!("[flows][inference-probe] role={role} managed provider registered no probe");
return Ok(());
};
+ let provider_context = crate::inference::registry::ProviderContext {
+ config: config.clone(),
+ egress: crate::security::egress::EgressLedger::none(),
+ };
let model_id = entry
.models
.as_ref()
- .and_then(|models| (models.resolve_model)(role, config, None))
+ .and_then(|models| (models.resolve_model)(role, &provider_context, None))
.or_else(|| {
config
.default_model
@@ -518,7 +522,11 @@ pub async fn probe_inference_readiness(
.map(str::to_string)
})
.ok_or_else(|| format!("provider {PROVIDER_OPENHUMAN} has no model for role {role}"))?;
- let result = probe(role.to_string(), model_id.clone(), config.clone()).await;
+ let result = probe(tinyinference_llm::providers::overrides::ReadinessRequest {
+ role: role.to_string(),
+ model: model_id.clone(),
+ context: provider_context,
+ }).await;
log::debug!(
"[flows][inference-probe] role={role} model={model_id} probe result: {}",
if result.is_ok() { "ready" } else { "not ready" }
diff --git a/crates/openhuman-core/src/inference/provider/factory/managed_backend.rs b/crates/openhuman-core/src/inference/provider/factory/managed_backend.rs
index 2181fe4aee..3aea72c132 100644
--- a/crates/openhuman-core/src/inference/provider/factory/managed_backend.rs
+++ b/crates/openhuman-core/src/inference/provider/factory/managed_backend.rs
@@ -67,7 +67,7 @@ pub(super) fn managed_chat_model_for(
return Err(managed_unavailable_error());
};
let model = match entry.models.as_ref() {
- Some(models) => (models.resolve_model)(role, config, model_override),
+ Some(models) => (models.resolve_model)(role, &crate::inference::registry::ProviderContext { config: config.clone(), egress: egress.clone() }, model_override),
None => model_override
.or(config.default_model.as_deref())
.map(str::trim)
@@ -86,14 +86,15 @@ pub(super) fn managed_chat_model_for(
config: config.clone(),
egress: egress.clone(),
};
- (entry.build)(&crate::inference::registry::ChatModelRequest {
+ let built = (entry.build)(&crate::inference::registry::ChatModelRequest {
role,
model_override: Some(&model),
default_model,
native_tool_calling,
thread_id,
context: &context,
- })
+ })?;
+ Ok(built)
}
/// Construct a managed turn model with the explicit OpenHuman thread attached.
@@ -188,12 +189,13 @@ pub(crate) fn make_openhuman_backend_model_for_options(
config: config.clone(),
egress,
};
- (entry.build)(&crate::inference::registry::ChatModelRequest {
+ let built = (entry.build)(&crate::inference::registry::ChatModelRequest {
role: "delegate",
model_override: Some(model),
default_model: Some(model),
native_tool_calling: true,
thread_id,
context: &context,
- })
+ })?;
+ Ok(built)
}