add overrides.disable_signals to skip CPU-heavy signal analysis (#906)

config/plano_config_schema.yaml

@@ -278,6 +278,9 @@ properties:
         type: boolean
       use_agent_orchestrator:
         type: boolean
+      disable_signals:
+        type: boolean
+        description: "Disable agentic signal analysis (frustration, repetition, escalation, etc.) on LLM responses to save CPU. Default false."
       upstream_connect_timeout:
         type: string
         description: "Connect timeout for upstream provider clusters (e.g., '5s', '10s'). Default is '5s'."

crates/brightstaff/src/app_state.rs

@@ -24,4 +24,7 @@ pub struct AppState {
     /// Shared HTTP client for upstream LLM requests (connection pooling / keep-alive).
     pub http_client: reqwest::Client,
     pub filter_pipeline: Arc<FilterPipeline>,
+    /// When false, agentic signal analysis is skipped on LLM responses to save CPU.
+    /// Controlled by `overrides.disable_signals` in plano config.
+    pub signals_enabled: bool,
 }

crates/brightstaff/src/handlers/llm/mod.rs

@@ -143,6 +143,7 @@ async fn llm_chat_inner(
         &request_path,
         &state.model_aliases,
         &state.llm_providers,
+        state.signals_enabled,
     )
     .await
     {
@@ -408,6 +409,7 @@ async fn parse_and_validate_request(
     request_path: &str,
     model_aliases: &Option<HashMap<String, ModelAlias>>,
     llm_providers: &Arc<RwLock<LlmProviders>>,
+    signals_enabled: bool,
 ) -> Result<PreparedRequest, Response<BoxBody<Bytes, hyper::Error>>> {
     let raw_bytes = request
         .collect()
@@ -486,7 +488,11 @@ async fn parse_and_validate_request(
     let user_message_preview = client_request
         .get_recent_user_message()
         .map(|msg| truncate_message(&msg, 50));
-    let messages_for_signals = Some(client_request.get_messages());
+    let messages_for_signals = if signals_enabled {
+        Some(client_request.get_messages())
+    } else {
+        None
+    };
 
     // Set the upstream model name and strip routing metadata
     client_request.set_model(model_name_only.clone());

crates/brightstaff/src/main.rs

@@ -328,6 +328,8 @@ async fn init_app_state(
         .as_ref()
         .and_then(|tracing| tracing.span_attributes.clone());
 
+    let signals_enabled = !overrides.disable_signals.unwrap_or(false);
+
     Ok(AppState {
         orchestrator_service,
         model_aliases: config.model_aliases.clone(),
@@ -339,6 +341,7 @@ async fn init_app_state(
         span_attributes,
         http_client: reqwest::Client::new(),
         filter_pipeline,
+        signals_enabled,
     })
 }
 

crates/common/src/configuration.rs

@@ -234,6 +234,7 @@ pub struct Overrides {
     pub llm_routing_model: Option<String>,
     pub agent_orchestration_model: Option<String>,
     pub orchestrator_model_context_length: Option<usize>,
+    pub disable_signals: Option<bool>,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
@@ -750,4 +751,29 @@ mod test {
         assert!(model_ids.contains(&"openai-gpt4".to_string()));
         assert!(!model_ids.contains(&"plano-orchestrator".to_string()));
     }
+
+    #[test]
+    fn test_overrides_disable_signals_default_none() {
+        let overrides = super::Overrides::default();
+        assert_eq!(overrides.disable_signals, None);
+    }
+
+    #[test]
+    fn test_overrides_disable_signals_deserialize() {
+        let yaml = r#"
+disable_signals: true
+"#;
+        let overrides: super::Overrides = serde_yaml::from_str(yaml).unwrap();
+        assert_eq!(overrides.disable_signals, Some(true));
+
+        let yaml_false = r#"
+disable_signals: false
+"#;
+        let overrides: super::Overrides = serde_yaml::from_str(yaml_false).unwrap();
+        assert_eq!(overrides.disable_signals, Some(false));
+
+        let yaml_missing = "{}";
+        let overrides: super::Overrides = serde_yaml::from_str(yaml_missing).unwrap();
+        assert_eq!(overrides.disable_signals, None);
+    }
 }

docs/source/resources/includes/plano_config_full_reference.yaml

@@ -173,6 +173,9 @@ overrides:
   llm_routing_model: Plano-Orchestrator
   # Model used for agent orchestration (must be listed in model_providers)
   agent_orchestration_model: Plano-Orchestrator
+  # Disable agentic signal analysis (frustration, repetition, escalation, etc.)
+  # on LLM responses to save CPU. Default: false.
+  disable_signals: false
 
 # Model affinity — pin routing decisions for agentic loops
 routing:

docs/source/resources/includes/plano_config_full_reference_rendered.yaml

@@ -170,6 +170,7 @@ model_providers:
   provider_interface: plano
 overrides:
   agent_orchestration_model: Plano-Orchestrator
+  disable_signals: false
   llm_routing_model: Plano-Orchestrator
   optimize_context_window: true
   prompt_target_intent_matching_threshold: 0.7