feat: first cut at openai compatible server

src/database.rs

@@ -51,6 +51,7 @@ pub struct AppPreferences {
     pub directional_steering_attn: Option<f32>,
     pub simulated_used_memory_gib: Option<i64>,
     pub expert_profile_path: Option<String>,
+    pub endpoint_port: i32,
 }
 
 impl Default for AppPreferences {
@@ -90,6 +91,7 @@ impl Default for AppPreferences {
             directional_steering_attn: None,
             simulated_used_memory_gib: None,
             expert_profile_path: None,
+            endpoint_port: 4000,
         }
     }
 }
@@ -234,6 +236,7 @@ struct PreferenceChanges<'a> {
     directional_steering_attn: Option<f32>,
     simulated_used_memory_gib: Option<i64>,
     expert_profile_path: Option<&'a str>,
+    endpoint_port: i32,
 }
 
 #[derive(Clone, Debug, Identifiable, Queryable, Selectable)]
@@ -360,6 +363,7 @@ impl Database {
         &mut self,
         selected_model: &str,
         idle_timeout_minutes: i32,
+        endpoint_port: i32,
         generation: &GenerationPreferences,
         runtime: &RuntimePreferences,
     ) -> Result<AppPreferences, String> {
@@ -367,6 +371,9 @@ impl Database {
         let model = crate::model::ModelChoice::from_id(selected_model)
             .ok_or_else(|| format!("Unsupported model: {selected_model}"))?;
         runtime.validate(model)?;
+        if !(1..=65_535).contains(&endpoint_port) {
+            return Err("Endpoint port must be between 1 and 65535.".into());
+        }
         let execution = &runtime.execution;
         let speculative = &runtime.speculative;
         let (ssd_cache_experts, ssd_cache_gib) = match runtime.ssd.cache {
@@ -416,6 +423,7 @@ impl Database {
                     .simulated_used_memory_gib
                     .map(|value| value as i64),
                 expert_profile_path: runtime.diagnostics.expert_profile_path.as_deref(),
+                endpoint_port,
             })
             .returning(AppPreferences::as_returning())
             .get_result(&mut self.connection)
@@ -572,6 +580,7 @@ mod tests {
         assert_eq!(preferences.selected_model, "deepseek-v4-flash");
         assert!(!preferences.dspark_enabled);
         assert_eq!(preferences.idle_timeout_minutes, 10);
+        assert_eq!(preferences.endpoint_port, 4000);
         let generation = GenerationPreferences::default();
         let runtime = RuntimePreferences::default();
         assert!(
@@ -579,6 +588,7 @@ mod tests {
                 .update_preferences(
                     "glm-5.2",
                     30,
+                    4000,
                     &generation,
                     &RuntimePreferences {
                         speculative: SpeculativePreferences {
@@ -592,7 +602,7 @@ mod tests {
         );
         assert!(
             database
-                .update_preferences("deepseek-v4-flash", 0, &generation, &runtime,)
+                .update_preferences("deepseek-v4-flash", 0, 4000, &generation, &runtime,)
                 .is_err()
         );
         let generation = GenerationPreferences {
@@ -630,7 +640,7 @@ mod tests {
             ..RuntimePreferences::default()
         };
         database
-            .update_preferences("glm-5.2", 30, &generation, &runtime)
+            .update_preferences("glm-5.2", 30, 4567, &generation, &runtime)
             .unwrap();
 
         let project = database.create_project("DS4", "/tmp/ds4").unwrap();
@@ -652,6 +662,7 @@ mod tests {
         let preferences = reopened.load_preferences().unwrap();
         assert_eq!(preferences.selected_model, "glm-5.2");
         assert_eq!(preferences.idle_timeout_minutes, 30);
+        assert_eq!(preferences.endpoint_port, 4567);
         assert_eq!(preferences.generation().unwrap(), generation);
         assert_eq!(preferences.runtime().unwrap(), runtime);
         drop(reopened);