Feat/modelmanager

crates/edgen_core/src/settings.rs

@@ -95,6 +95,30 @@ fn build_config_file_path() -> PathBuf {
     config_dir.join(Path::new(&filename))
 }
 
+/// Helper to get the chat completions model directory.
+pub async fn chat_completions_dir() -> String {
+    SETTINGS
+        .read()
+        .await
+        .read()
+        .await
+        .chat_completions_models_dir
+        .trim()
+        .to_string()
+}
+
+/// Helper to get the audio transcriptions model directory.
+pub async fn audio_transcriptions_dir() -> String {
+    SETTINGS
+        .read()
+        .await
+        .read()
+        .await
+        .audio_transcriptions_models_dir
+        .trim()
+        .to_string()
+}
+
 #[derive(Error, Debug, Serialize)]
 pub enum SettingsError {
     #[error("failed to read the settings file: {0}")]

crates/edgen_server/Cargo.toml

@@ -43,3 +43,4 @@ testcontainers = "0.15.0"
 
 [dev-dependencies]
 levenshtein = "1.0.5"
+tempfile = { workspace = true }

crates/edgen_server/src/lib.rs

@@ -20,7 +20,6 @@ use std::process::exit;
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
 
-use axum::Router;
 use tower_http::cors::CorsLayer;
 
 use futures::executor::block_on;
@@ -43,7 +42,9 @@ pub mod error;
 pub mod graceful_shutdown;
 mod llm;
 mod model;
+pub mod model_man;
 pub mod openai_shim;
+mod routes;
 pub mod status;
 pub mod util;
 mod whisper;
@@ -189,32 +190,7 @@ async fn run_server(args: &cli::Serve) -> Result<bool, error::EdgenError> {
     )
     .await;
 
-    let http_app = Router::new()
-        // -- AI endpoints -----------------------------------------------------
-        // ---- Chat -----------------------------------------------------------
-        .route(
-            "/v1/chat/completions",
-            axum::routing::post(openai_shim::chat_completions),
-        )
-        // ---- Audio ----------------------------------------------------------
-        .route(
-            "/v1/audio/transcriptions",
-            axum::routing::post(openai_shim::create_transcription),
-        )
-        // -- AI status endpoints ----------------------------------------------
-        // ---- Chat -----------------------------------------------------------
-        .route(
-            "/v1/chat/completions/status",
-            axum::routing::get(status::chat_completions_status),
-        )
-        // ---- Audio ----------------------------------------------------------
-        .route(
-            "/v1/audio/transcriptions/status",
-            axum::routing::get(status::audio_transcriptions_status),
-        )
-        // -- Miscellaneous services -------------------------------------------
-        .route("/v1/misc/version", axum::routing::get(misc::edgen_version))
-        .layer(CorsLayer::permissive());
+    let http_app = routes::routes().layer(CorsLayer::permissive());
 
     let uri_vector = if !args.uri.is_empty() {
         info!("Overriding default URI");

crates/edgen_server/src/misc.rs

@@ -36,7 +36,7 @@ pub struct Version {
     build: String,
 }
 
-/// GET `/v1/version`: returns the current version of edgend.
+/// GET `/v1/misc/version`: returns the current version of edgend.
 ///
 /// The version is returned as json value with major, minor and patch as integer
 /// and build as string (which may be empty).