vLLM priority scheduling for agents

Thread request priority through the API call chain to vLLM's
priority scheduler. Lower value = higher priority, with preemption.

Priority is set per-agent in the .agent header:
- interactive (runner): 0 (default, highest)
- surface-observe: 1 (near-realtime, watches conversation)
- all other agents: 10 (batch, default if not specified)

Requires vLLM started with --scheduling-policy priority.

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/subconscious/agents/surface-observe.agent

@@ -1,4 +1,4 @@
-{"agent":"surface-observe","query":"","count":1,"bail":"bail-no-competing.sh"}
+{"agent":"surface-observe","query":"","count":1,"priority":1,"bail":"bail-no-competing.sh"}
 
 === PROMPT phase:surface ===
 

src/subconscious/api.rs

@@ -35,6 +35,7 @@ pub async fn call_api_with_tools(
     prompts: &[String],
     phases: &[String],
     temperature: Option<f32>,
+    priority: i32,
     tools: &[String],
     bail_fn: Option<&(dyn Fn(usize) -> Result<(), String> + Sync)>,
     log: &dyn Fn(&str),
@@ -82,6 +83,7 @@ pub async fn call_api_with_tools(
                 &ui_tx,
                 &reasoning,
                 temperature,
+                Some(priority),
             ).await {
                 Ok((msg, usage)) => {
                     msg_opt = Some(msg);
@@ -233,6 +235,7 @@ pub fn call_api_with_tools_sync(
     prompts: &[String],
     phases: &[String],
     temperature: Option<f32>,
+    priority: i32,
     tools: &[String],
     bail_fn: Option<&(dyn Fn(usize) -> Result<(), String> + Sync)>,
     log: &(dyn Fn(&str) + Sync),
@@ -244,7 +247,7 @@ pub fn call_api_with_tools_sync(
                 .build()
                 .map_err(|e| format!("tokio runtime: {}", e))?;
             rt.block_on(
-                    call_api_with_tools(agent, prompts, phases, temperature, tools, bail_fn, log)
+                    call_api_with_tools(agent, prompts, phases, temperature, priority, tools, bail_fn, log)
             )
         }).join().unwrap()
     })

src/subconscious/defs.rs

@@ -46,6 +46,9 @@ pub struct AgentDef {
     pub chunk_size: Option<usize>,
     pub chunk_overlap: Option<usize>,
     pub temperature: Option<f32>,
+    /// vLLM scheduling priority (lower = higher priority).
+    /// 0 = interactive, 1 = near-realtime, 10 = batch (default).
+    pub priority: i32,
     /// Bail check command — run between steps with pid file path as $1,
     /// cwd = state dir. Non-zero exit = stop the pipeline.
     pub bail: Option<String>,
@@ -75,6 +78,9 @@ struct AgentHeader {
     /// LLM temperature override
     #[serde(default)]
     temperature: Option<f32>,
+    /// vLLM scheduling priority (lower = higher priority, default 10 = batch)
+    #[serde(default = "default_priority")]
+    priority: i32,
     /// Bail check command — run between steps with pid file path as $1,
     /// cwd = state dir. Non-zero exit = stop the pipeline.
     #[serde(default)]
@@ -82,6 +88,7 @@ struct AgentHeader {
 }
 
 fn default_model() -> String { "sonnet".into() }
+fn default_priority() -> i32 { 10 }
 
 /// Parse an agent file: first line is JSON config, rest is the prompt(s).
 /// Multiple prompts are separated by `=== PROMPT [phase:name] ===` lines.
@@ -149,6 +156,7 @@ fn parse_agent_file(content: &str) -> Option<AgentDef> {
         chunk_size: header.chunk_size,
         chunk_overlap: header.chunk_overlap,
         temperature: header.temperature,
+        priority: header.priority,
         bail: header.bail,
     })
 }

src/subconscious/llm.rs

@@ -22,7 +22,7 @@ pub(crate) fn call_simple(caller: &str, prompt: &str) -> Result<String, String>
 
     let prompts = vec![prompt.to_string()];
     let phases = vec![];
-    super::api::call_api_with_tools_sync(caller, &prompts, &phases, None, &[], None, &log)
+    super::api::call_api_with_tools_sync(caller, &prompts, &phases, None, 10, &[], None, &log)
 }
 
 /// Call a model using an agent definition's configuration (multi-step).
@@ -34,7 +34,7 @@ pub(crate) fn call_for_def_multi(
     bail_fn: Option<&(dyn Fn(usize) -> Result<(), String> + Sync)>,
     log: &(dyn Fn(&str) + Sync),
 ) -> Result<String, String> {
-    super::api::call_api_with_tools_sync(&def.agent, prompts, phases, def.temperature, &def.tools, bail_fn, log)
+    super::api::call_api_with_tools_sync(&def.agent, prompts, phases, def.temperature, def.priority, &def.tools, bail_fn, log)
 }