agent: add sampling parameters (temperature, top_p, top_k)

Move temperature from a per-call parameter to an Agent field,
add top_p and top_k. All three are sent to the API via a new
SamplingParams struct, displayed on the F5 thalamus screen.

Defaults: temperature=0.6, top_p=0.95, top_k=20 (Qwen3.5 defaults).

Also adds top_p and top_k to ChatRequest so they're sent in the
API payload. Previously only temperature was sent.

UI controls for adjusting these at runtime are not yet implemented.

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

src/agent/api/mod.rs

@@ -29,6 +29,14 @@ impl Drop for AbortOnDrop {
     }
 }
 
+/// Sampling parameters for model generation.
+#[derive(Clone, Copy)]
+pub struct SamplingParams {
+    pub temperature: f32,
+    pub top_p: f32,
+    pub top_k: u32,
+}
+
 // ─────────────────────────────────────────────────────────────
 //  Stream events — yielded by backends, consumed by the runner
 // ─────────────────────────────────────────────────────────────
@@ -93,7 +101,7 @@ impl ApiClient {
         tools: Option<&[ToolDef]>,
         ui_tx: &UiSender,
         reasoning_effort: &str,
-        temperature: Option<f32>,
+        sampling: SamplingParams,
         priority: Option<i32>,
     ) -> (mpsc::UnboundedReceiver<StreamEvent>, AbortOnDrop) {
         let (tx, rx) = mpsc::unbounded_channel();
@@ -110,7 +118,7 @@ impl ApiClient {
             let result = openai::stream_events(
                 &client, &base_url, &api_key, &model,
                 &messages, tools.as_deref(), &tx, &ui_tx,
-                &reasoning_effort, temperature, priority,
+                &reasoning_effort, sampling, priority,
             ).await;
             if let Err(e) = result {
                 let _ = tx.send(StreamEvent::Error(e.to_string()));
@@ -126,11 +134,11 @@ impl ApiClient {
         tools: Option<&[ToolDef]>,
         ui_tx: &UiSender,
         reasoning_effort: &str,
-        temperature: Option<f32>,
+        sampling: SamplingParams,
         priority: Option<i32>,
     ) -> Result<(Message, Option<Usage>)> {
         // Use the event stream and accumulate into a message.
-        let (mut rx, _handle) = self.start_stream(messages, tools, ui_tx, reasoning_effort, temperature, priority);
+        let (mut rx, _handle) = self.start_stream(messages, tools, ui_tx, reasoning_effort, sampling, priority);
         let mut content = String::new();
         let mut tool_calls: Vec<ToolCall> = Vec::new();
         let mut usage = None;

src/agent/api/openai.rs

@@ -26,7 +26,7 @@ pub(super) async fn stream_events(
     tx: &mpsc::UnboundedSender<StreamEvent>,
     ui_tx: &UiSender,
     reasoning_effort: &str,
-    temperature: Option<f32>,
+    sampling: super::SamplingParams,
     priority: Option<i32>,
 ) -> Result<()> {
     let request = ChatRequest {
@@ -35,7 +35,9 @@ pub(super) async fn stream_events(
         tool_choice: tools.map(|_| "auto".to_string()),
         tools: tools.map(|t| t.to_vec()),
         max_tokens: Some(16384),
-        temperature: Some(temperature.unwrap_or(0.6)),
+        temperature: Some(sampling.temperature),
+        top_p: Some(sampling.top_p),
+        top_k: Some(sampling.top_k),
         stream: Some(true),
         reasoning: if reasoning_effort != "none" && reasoning_effort != "default" {
             Some(ReasoningConfig {

src/agent/api/types.rs

@@ -95,6 +95,10 @@ pub struct ChatRequest {
     #[serde(skip_serializing_if = "Option::is_none")]
     pub temperature: Option<f32>,
     #[serde(skip_serializing_if = "Option::is_none")]
+    pub top_p: Option<f32>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub top_k: Option<u32>,
+    #[serde(skip_serializing_if = "Option::is_none")]
     pub stream: Option<bool>,
     /// OpenRouter reasoning control. Send both formats for compatibility:
     /// - reasoning.enabled (older format, still seen in examples)

src/agent/mod.rs

@@ -77,6 +77,10 @@ pub struct Agent {
     last_prompt_tokens: u32,
     /// Current reasoning effort level ("none", "low", "high").
     pub reasoning_effort: String,
+    /// Sampling parameters — adjustable at runtime from the thalamus screen.
+    pub temperature: f32,
+    pub top_p: f32,
+    pub top_k: u32,
     /// Persistent conversation log — append-only record of all messages.
     conversation_log: Option<ConversationLog>,
     /// BPE tokenizer for token counting (cl100k_base — close enough
@@ -137,6 +141,9 @@ impl Agent {
             tool_defs,
             last_prompt_tokens: 0,
             reasoning_effort: "none".to_string(),
+            temperature: 0.6,
+            top_p: 0.95,
+            top_k: 20,
             conversation_log,
             tokenizer,
             context,
@@ -288,12 +295,17 @@ impl Agent {
             let (mut rx, _stream_guard) = {
                 let me = agent.lock().await;
                 let api_messages = me.assemble_api_messages();
+                let sampling = api::SamplingParams {
+                    temperature: me.temperature,
+                    top_p: me.top_p,
+                    top_k: me.top_k,
+                };
                 me.client.start_stream(
                     &api_messages,
                     Some(&me.tool_defs),
                     ui_tx,
                     &me.reasoning_effort,
-                    None,
+                    sampling,
                     None,
                 )
             };

src/subconscious/api.rs

@@ -76,12 +76,17 @@ pub async fn call_api_with_tools(
         let mut msg_opt = None;
         let mut usage_opt = None;
         for attempt in 0..5 {
+            let sampling = crate::agent::api::SamplingParams {
+                temperature: temperature.unwrap_or(0.6),
+                top_p: 0.95,
+                top_k: 20,
+            };
             match client.chat_completion_stream_temp(
                 &messages,
                 Some(&tool_defs),
                 &ui_tx,
                 &reasoning,
-                temperature,
+                sampling,
                 Some(priority),
             ).await {
                 Ok((msg, usage)) => {

src/user/mod.rs

@@ -273,6 +273,9 @@ pub struct App {
     pub(crate) needs_assistant_marker: bool,
     pub running_processes: u32,
     pub reasoning_effort: String,
+    pub temperature: f32,
+    pub top_p: f32,
+    pub top_k: u32,
     pub(crate) active_tools: crate::user::ui_channel::SharedActiveTools,
     pub(crate) active_pane: ActivePane,
     pub textarea: tui_textarea::TextArea<'static>,
@@ -310,6 +313,9 @@ impl App {
             turn_started: None, call_started: None, call_timeout_secs: 60,
             needs_assistant_marker: false, running_processes: 0,
             reasoning_effort: "none".to_string(),
+            temperature: 0.6,
+            top_p: 0.95,
+            top_k: 20,
             active_tools, active_pane: ActivePane::Conversation,
             textarea: new_textarea(vec![String::new()]),
             input_history: Vec::new(), history_index: None,

src/user/thalamus.rs

@@ -48,6 +48,14 @@ impl App {
         }
         lines.push(Line::raw(""));
 
+        // Sampling parameters
+        lines.push(Line::styled("── Sampling ──", section));
+        lines.push(Line::raw(""));
+        lines.push(Line::raw(format!("  temperature: {:.2}", self.temperature)));
+        lines.push(Line::raw(format!("  top_p:       {:.2}", self.top_p)));
+        lines.push(Line::raw(format!("  top_k:       {}", self.top_k)));
+        lines.push(Line::raw(""));
+
         // Channel status from cached data
         lines.push(Line::styled("── Channels ──", section));
         lines.push(Line::raw(""));