fix: model repo subdir layout, no-window engine launch, deploy progress, engine path auto-repair

README.md

@@ -4,7 +4,7 @@
 
 ## 当前功能
 
-- **模型库**：启动时自动扫描模型目录（可手动重新扫描），支持导入本地 GGUF、添加 OpenAI 兼容在线 API 模型（聊天时可直接选用）
+- **模型管理**：启动时自动扫描模型目录（可手动重新扫描），支持导入本地 GGUF、添加 OpenAI 兼容在线 API 模型；本地模型可一键「部署」后台加载，部署成功后对话页可直接选用
 - **模型广场**：搜索 Hugging Face / ModelScope 上的 GGUF 模型，按量化版本一键下载（ModelScope 文件自动附带 SHA256 校验），下载进度实时显示
 - **聊天**：流式输出、Markdown/代码高亮、采样参数调节、多会话管理；本地模型与在线 API 模型统一入口
 - **本地 API 服务**：OpenAI 兼容端点（/v1/models、/v1/chat/completions、/v1/embeddings），仅本机监听，可选 API Key

apps/desktop/src/commands.rs

@@ -62,6 +62,22 @@ pub struct ChatErrorEvent {
     pub message: String,
 }
 
+#[derive(Serialize, Clone)]
+pub struct EngineDeployEvent {
+    pub model_id: String,
+    pub file_name: String,
+    pub state: String,
+    pub message: Option<String>,
+}
+
+#[derive(Serialize, Clone)]
+pub struct EngineDeployProgressEvent {
+    pub model_id: String,
+    pub file_name: String,
+    pub percent: u32,
+    pub stage: String,
+}
+
 #[derive(Deserialize, Clone)]
 pub struct DownloadPayload {
     pub url: String,
@@ -196,6 +212,7 @@ pub fn get_messages(state: State<'_, App>, conversation_id: String) -> Result<Ve
 
 #[tauri::command]
 pub async fn engine_start(
+    app: AppHandle,
     state: State<'_, App>,
     model_id: String,
     params: ChatParams,
@@ -237,13 +254,16 @@ pub async fn engine_start(
     };
     engine.start(cfg).await.map_err(|e| e.to_string())?;
     *base.write().await = engine.base_url().await;
-    Ok(engine.status().await)
+    let status = engine.status().await;
+    let _ = app.emit("engine://status", status.clone());
+    Ok(status)
 }
 
 #[tauri::command]
-pub async fn engine_stop(state: State<'_, App>) -> Result<(), String> {
+pub async fn engine_stop(app: AppHandle, state: State<'_, App>) -> Result<(), String> {
     state.engine.stop().await.map_err(|e| e.to_string())?;
     *state.engine_base.write().await = None;
+    let _ = app.emit("engine://status", state.engine.status().await);
     Ok(())
 }
 
@@ -254,6 +274,159 @@ pub async fn engine_status(
     Ok(state.engine.status().await)
 }
 
+/// 后台部署（加载）本地模型：启动 llama-server 并等待就绪，通过事件广播进度。
+#[tauri::command]
+pub async fn deploy_model(
+    app: AppHandle,
+    state: State<'_, App>,
+    model_id: String,
+    params: ChatParams,
+) -> Result<(), String> {
+    let core = state.core.clone();
+    let engine = state.engine.clone();
+    let base = state.engine_base.clone();
+
+    let (model, engine_bin) = {
+        let db = core.db.lock().unwrap();
+        let model = models_db::get(&db, &model_id)
+            .map_err(|e| e.to_string())?
+            .ok_or_else(|| "model not found".to_string())?;
+        if model.kind != "local" {
+            return Err("在线 API 模型无需部署".into());
+        }
+        let bin = settings_db::get(&db, "engine_bin")
+            .map_err(|e| e.to_string())?
+            .unwrap_or_default();
+        (model, bin)
+    };
+
+    let file_name = model.file_name.clone();
+    let emit_deploy = |state_name: &str, message: Option<String>| {
+        let _ = app.emit(
+            "engine://deploy",
+            EngineDeployEvent {
+                model_id: model_id.clone(),
+                file_name: file_name.clone(),
+                state: state_name.into(),
+                message,
+            },
+        );
+    };
+
+    let current = engine.status().await;
+    if current.running {
+        if current.model.as_deref() == Some(file_name.as_str()) {
+            emit_deploy("ready", None);
+            return Ok(());
+        }
+        // 部署另一个模型前，先停掉当前引擎
+        let _ = engine.stop().await;
+        *base.write().await = None;
+        let _ = app.emit("engine://status", engine.status().await);
+    }
+
+    let bin = PathBuf::from(engine_bin);
+    if !bin.exists() {
+        emit_deploy(
+            "error",
+            Some(format!(
+                "引擎文件不存在：{}，请到设置页配置 llama-server 路径",
+                bin.display()
+            )),
+        );
+        return Ok(());
+    }
+
+    emit_deploy("loading", None);
+    let app2 = app.clone();
+    let core2 = core.clone();
+    let engine2 = engine.clone();
+    let base2 = base.clone();
+    let model_id2 = model_id.clone();
+    let file_name2 = file_name.clone();
+
+    tokio::spawn(async move {
+        let cfg = EngineConfig {
+            binary_path: bin,
+            model_path: PathBuf::from(model.file_path),
+            host: "127.0.0.1".into(),
+            ctx_size: params.ctx_size,
+            ngl: params.ngl,
+            threads: None,
+            log_file: core2.logs_dir.join(format!("engine-{}.log", model.file_name)),
+        };
+        // 后台轮询引擎日志，解析模型加载进度并广播
+        let log_path = cfg.log_file.clone();
+        let progress_app = app2.clone();
+        let progress_model = model_id2.clone();
+        let progress_name = file_name2.clone();
+        let progress_task = tokio::spawn(async move {
+            let mut last_size = 0u64;
+            loop {
+                if let Ok(meta) = tokio::fs::metadata(&log_path).await {
+                    let size = meta.len();
+                    if size != last_size {
+                        last_size = size;
+                        if let Ok(bytes) = tokio::fs::read(&log_path).await {
+                            let text = String::from_utf8_lossy(&bytes);
+                            let (percent, stage) = estimate_load_progress(&text);
+                            let _ = progress_app.emit(
+                                "engine://deploy-progress",
+                                EngineDeployProgressEvent {
+                                    model_id: progress_model.clone(),
+                                    file_name: progress_name.clone(),
+                                    percent,
+                                    stage: stage.to_string(),
+                                },
+                            );
+                        }
+                    }
+                }
+                tokio::time::sleep(std::time::Duration::from_millis(250)).await;
+            }
+        });
+
+        match engine2.start(cfg).await {
+            Ok(()) => {
+                progress_task.abort();
+                *base2.write().await = engine2.base_url().await;
+                let _ = app2.emit(
+                    "engine://deploy",
+                    EngineDeployEvent {
+                        model_id: model_id2.clone(),
+                        file_name: file_name2.clone(),
+                        state: "ready".into(),
+                        message: None,
+                    },
+                );
+                let _ = app2.emit(
+                    "engine://deploy-progress",
+                    EngineDeployProgressEvent {
+                        model_id: model_id2.clone(),
+                        file_name: file_name2.clone(),
+                        percent: 100,
+                        stage: "服务已就绪".into(),
+                    },
+                );
+            }
+            Err(e) => {
+                progress_task.abort();
+                let _ = app2.emit(
+                    "engine://deploy",
+                    EngineDeployEvent {
+                        model_id: model_id2,
+                        file_name: file_name2,
+                        state: "error".into(),
+                        message: Some(e.to_string()),
+                    },
+                );
+            }
+        }
+        let _ = app2.emit("engine://status", engine2.status().await);
+    });
+    Ok(())
+}
+
 #[tauri::command]
 pub async fn chat_send(
     app: AppHandle,
@@ -417,7 +590,12 @@ pub fn download_enqueue(
 ) -> Result<String, String> {
     let id = uuid::Uuid::new_v4().to_string();
     let core = state.core.clone();
-    let dest = core.models_dir.join(&payload.file_name);
+    // 模型按 <models_dir>/<owner>/<repo>/<file> 组织，不直接放模型目录根
+    let subdir = derive_repo_subdir(&payload);
+    let dest_dir = core.models_dir.join(&subdir);
+    std::fs::create_dir_all(&dest_dir)
+        .map_err(|e| format!("failed to create model directory {}: {e}", dest_dir.display()))?;
+    let dest = dest_dir.join(&payload.file_name);
     let opts = xianren_download::DownloadOptions {
         url: payload.url.clone(),
         dest,
@@ -852,6 +1030,54 @@ async fn drain_stream_and_persist(
     );
 }
 
+/// 根据 repo_id 或 URL 推断下载模型应存放的子目录（owner/repo）。
+fn derive_repo_subdir(payload: &DownloadPayload) -> String {
+    if let Some(repo_id) = &payload.repo_id {
+        let parts: Vec<&str> = repo_id.split('/').collect();
+        if parts.len() >= 2 && !parts[0].trim().is_empty() && !parts[1].trim().is_empty() {
+            return format!("{}/{}", parts[0].trim(), parts[1].trim());
+        }
+    }
+    if let Some(rest) = payload.url.split("://").nth(1) {
+        let segments: Vec<&str> = rest.split('/').collect();
+        let start = if segments.get(1) == Some(&"models") {
+            2
+        } else {
+            1
+        };
+        if segments.len() >= start + 2
+            && !segments[start].trim().is_empty()
+            && !segments[start + 1].trim().is_empty()
+        {
+            return format!("{}/{}", segments[start].trim(), segments[start + 1].trim());
+        }
+    }
+    "manual".to_string()
+}
+
+/// 根据引擎日志内容估算模型加载进度。
+fn estimate_load_progress(log: &str) -> (u32, &'static str) {
+    if log.contains("listening on http")
+        || log.contains("server is listening")
+        || log.contains("HTTP server listening")
+    {
+        (100, "服务已就绪")
+    } else if log.contains("model loaded") || log.contains("llama_new_context_with_model") {
+        (90, "推理上下文就绪")
+    } else if log.contains("load_model: initializing") {
+        (70, "初始化推理上下文")
+    } else if log.contains("load_tensors") || log.contains("model size") {
+        (55, "加载模型权重")
+    } else if log.contains("load_model: loading model")
+        || log.contains("llama_model_load")
+        || log.contains("loading model")
+    {
+        (25, "读取模型文件")
+    } else {
+        (8, "启动引擎进程")
+    }
+}
+
 #[tauri::command]
 pub async fn server_start(
     state: State<'_, App>,

apps/desktop/src/lib.rs

@@ -7,7 +7,7 @@ use xianren_core::CoreApp;
 use xianren_core::models as models_db;
 use xianren_core::settings as settings_db;
 use xianren_engine::EngineManager;
-use std::path::PathBuf;
+use std::path::{Path, PathBuf};
 
 pub struct App {
     pub core: Arc<CoreApp>,
@@ -42,6 +42,34 @@ pub fn run() {
             // 启动时自动扫描模型目录
             let app_state = app.state::<App>();
             let db = app_state.core.db.lock().unwrap();
+            // 校验引擎路径：旧版本可能残留指向无效 exe 的配置（缺 DLL），自动修复
+            let configured = settings_db::get(&db, "engine_bin").ok().flatten();
+            let configured_ok = configured
+                .as_ref()
+                .map(|p| engine_binary_usable(PathBuf::from(p).as_path()))
+                .unwrap_or(false);
+            if !configured_ok {
+                let mut found: Option<PathBuf> = None;
+                for dir in ["cpu", "vulkan", "cuda"] {
+                    let candidate = app_state.core.engines_dir.join(dir).join("llama-server.exe");
+                    if engine_binary_usable(&candidate) {
+                        found = Some(candidate);
+                        break;
+                    }
+                }
+                if found.is_none() {
+                    let candidate = app_state.core.engines_dir.join("llama-server-cpu.exe");
+                    if engine_binary_usable(&candidate) {
+                        found = Some(candidate);
+                    }
+                }
+                if let Some(path) = found {
+                    let _ = settings_db::set(&db, "engine_bin", &path.to_string_lossy());
+                    tracing::info!(path = %path.display(), "engine binary auto-fixed");
+                } else {
+                    tracing::warn!("no usable llama-server binary found");
+                }
+            }
             let model_dir = settings_db::get(&db, "model_dir")
                 .ok()
                 .flatten()
@@ -81,6 +109,7 @@ pub fn run() {
             commands::engine_start,
             commands::engine_stop,
             commands::engine_status,
+            commands::deploy_model,
             commands::chat_send,
             commands::download_enqueue,
             commands::server_start,
@@ -91,3 +120,24 @@ pub fn run() {
         .run(tauri::generate_context!())
         .expect("error while running tauri application");
 }
+
+/// 判断引擎二进制是否可用：文件存在，且同目录包含 DLL（llama.cpp 发布版依赖同目录 DLL）。
+fn engine_binary_usable(path: &Path) -> bool {
+    if !path.is_file() {
+        return false;
+    }
+    let Some(parent) = path.parent() else {
+        return false;
+    };
+    std::fs::read_dir(parent)
+        .map(|entries| {
+            entries.flatten().any(|entry| {
+                entry
+                    .path()
+                    .extension()
+                    .map(|ext| ext.eq_ignore_ascii_case("dll"))
+                    .unwrap_or(false)
+            })
+        })
+        .unwrap_or(false)
+}

crates/engine/src/manager.rs

@@ -47,6 +47,9 @@ impl EngineManager {
             .unwrap_or_else(|| "unknown".to_string());
 
         let mut cmd = Command::new(&cfg.binary_path);
+        // Windows：以无窗口方式启动引擎（不弹出终端）
+        #[cfg(windows)]
+        cmd.creation_flags(0x0800_0000); // CREATE_NO_WINDOW
         cmd.arg("--model")
             .arg(&cfg.model_path)
             .arg("--host")

scripts/download-test-model.ps1

@@ -12,7 +12,10 @@ New-Item -ItemType Directory -Force -Path $OutDir | Out-Null
 
 $modelPath = "Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q4_k_m.gguf"
 $url = "$Endpoint/$modelPath"
-$outFile = Join-Path $OutDir "qwen2.5-0.5b-instruct-q4_k_m.gguf"
+$rel = "Qwen/Qwen2.5-0.5B-Instruct-GGUF"
+$targetDir = Join-Path $OutDir $rel
+New-Item -ItemType Directory -Force -Path $targetDir | Out-Null
+$outFile = Join-Path $targetDir "qwen2.5-0.5b-instruct-q4_k_m.gguf"
 
 Write-Host "Downloading $url"
 Write-Host "-> $outFile"

scripts/smoke-test.ps1

@@ -1,6 +1,6 @@
 param(
     [string]$Engine = "$env:APPDATA\XianrenStudio\engines\cpu\llama-server.exe",
-    [string]$Model = "$env:APPDATA\XianrenStudio\models\qwen2.5-0.5b-instruct-q4_k_m.gguf",
+    [string]$Model = "$env:APPDATA\XianrenStudio\models\Qwen\Qwen2.5-0.5B-Instruct-GGUF\qwen2.5-0.5b-instruct-q4_k_m.gguf",
     [int]$Port = 8088
 )
 

ui/src/App.tsx

@@ -1,7 +1,8 @@
-import { useEffect } from "react";
+import { useEffect, useState } from "react";
 import { NavLink, Route, Routes } from "react-router-dom";
-import { onEvent } from "./api";
+import { EngineDeployEvent, EngineDeployProgressEvent, onEvent } from "./api";
 import { useStore } from "./store";
+import Icon from "./components/Icon";
 import ModelsPage from "./pages/ModelsPage";
 import ChatPage from "./pages/ChatPage";
 import ModelPlazaPage from "./pages/ModelPlazaPage";
@@ -9,18 +10,31 @@ import ServerPage from "./pages/ServerPage";
 import SettingsPage from "./pages/SettingsPage";
 
 const navItems = [
-  { to: "/", label: "模型库", end: true },
-  { to: "/chat", label: "聊天" },
-  { to: "/plaza", label: "模型广场" },
-  { to: "/server", label: "本地服务" },
-  { to: "/settings", label: "设置" },
+  { to: "/chat", label: "对话", icon: "chat" },
+  { to: "/", label: "模型管理", icon: "box", end: true },
+  { to: "/plaza", label: "模型广场", icon: "plaza" },
+  { to: "/server", label: "服务管理", icon: "server" },
+  { to: "/settings", label: "设置", icon: "settings" },
 ];
 
 export default function App() {
+  const [collapsed, setCollapsed] = useState(
+    () => localStorage.getItem("xianren-nav-collapsed") === "1",
+  );
   const refreshModels = useStore((s) => s.refreshModels);
   const refreshEngine = useStore((s) => s.refreshEngine);
   const refreshServer = useStore((s) => s.refreshServer);
   const refreshConversations = useStore((s) => s.refreshConversations);
+  const setDeployState = useStore((s) => s.setDeployState);
+  const setDeployProgress = useStore((s) => s.setDeployProgress);
+
+  function toggleCollapsed() {
+    setCollapsed((v) => {
+      const next = !v;
+      localStorage.setItem("xianren-nav-collapsed", next ? "1" : "0");
+      return next;
+    });
+  }
 
   useEffect(() => {
     refreshModels();
@@ -31,43 +45,78 @@ export default function App() {
     const un2 = onEvent("server://status", () => refreshServer());
     const un3 = onEvent("download://done", () => refreshModels());
     const un4 = onEvent("models://updated", () => refreshModels());
+    const un5 = onEvent<EngineDeployEvent>("engine://deploy", (e) => {
+      setDeployState(e.model_id, e.state);
+      refreshEngine();
+    });
+    const un6 = onEvent<EngineDeployProgressEvent>("engine://deploy-progress", (e) => {
+      setDeployProgress(e.model_id, { percent: e.percent, stage: e.stage });
+    });
     return () => {
       un1.then((f) => f());
       un2.then((f) => f());
       un3.then((f) => f());
       un4.then((f) => f());
+      un5.then((f) => f());
+      un6.then((f) => f());
     };
-  }, [refreshModels, refreshEngine, refreshServer, refreshConversations]);
+  }, [
+    refreshModels,
+    refreshEngine,
+    refreshServer,
+    refreshConversations,
+    setDeployState,
+    setDeployProgress,
+  ]);
 
   return (
     <div className="flex h-screen overflow-hidden">
-      <aside className="flex w-56 flex-col gap-1 border-r border-border bg-panel p-3">
-        <div className="mb-4 flex items-center gap-2 px-2">
-          <div className="flex h-9 w-9 items-center justify-center rounded-xl bg-gradient-to-br from-accent to-accent-2 text-lg font-bold">
+      <aside
+        className={`flex flex-col gap-1 border-r border-border bg-panel p-2 transition-all duration-200 ${
+          collapsed ? "w-16" : "w-56"
+        }`}
+      >
+        <div
+          className={`mb-4 flex items-center ${collapsed ? "justify-center" : "gap-2 px-2"}`}
+        >
+          <div className="flex h-9 w-9 shrink-0 items-center justify-center rounded-xl bg-gradient-to-br from-accent to-accent-2 text-lg font-bold">
             仙
           </div>
-          <div>
-            <div className="text-sm font-semibold">仙人工作室</div>
-            <div className="text-xs text-slate-400">本地大模型工作台</div>
-          </div>
+          {!collapsed ? (
+            <div className="min-w-0">
+              <div className="truncate text-sm font-semibold">仙人工作室</div>
+              <div className="truncate text-xs text-slate-400">本地大模型工作台</div>
+            </div>
+          ) : null}
         </div>
         {navItems.map((item) => (
           <NavLink
             key={item.to}
             to={item.to}
             end={item.end}
+            title={item.label}
             className={({ isActive }) =>
-              `rounded-lg px-3 py-2 text-sm transition-colors ${
+              `flex items-center gap-3 rounded-lg px-3 py-2 text-sm transition-colors ${
+                collapsed ? "justify-center px-0" : ""
+              } ${
                 isActive
                   ? "bg-panel-2 text-white"
                   : "text-slate-300 hover:bg-panel-2/60"
               }`
             }
           >
-            {item.label}
+            <Icon name={item.icon} className="h-[18px] w-[18px] shrink-0" />
+            {!collapsed ? <span className="truncate">{item.label}</span> : null}
           </NavLink>
         ))}
-        <EngineChip />
+        <EngineChip collapsed={collapsed} />
+        <button
+          className="mt-1 flex items-center justify-center rounded-lg px-2 py-1.5 text-slate-400 hover:bg-panel-2/60 hover:text-slate-200"
+          onClick={toggleCollapsed}
+          title={collapsed ? "展开侧边栏" : "折叠侧边栏"}
+        >
+          <Icon name={collapsed ? "chevron-right" : "chevron-left"} />
+        </button>
       </aside>
       <main className="flex-1 overflow-hidden">
         <Routes>
@@ -82,9 +131,21 @@ export default function App() {
   );
 }
 
-function EngineChip() {
+function EngineChip({ collapsed }: { collapsed: boolean }) {
   const engine = useStore((s) => s.engine);
   const running = engine?.running ?? false;
+  if (collapsed) {
+    return (
+      <div
+        className="mt-auto flex justify-center rounded-lg py-2"
+        title={running ? `引擎运行中${engine?.model ? ` · ${engine.model}` : ""}` : "引擎未启动"}
+      >
+        <span
+          className={`h-2.5 w-2.5 rounded-full ${running ? "bg-emerald-400" : "bg-slate-500"}`}
+        />
+      </div>
+    );
+  }
   return (
     <div className="mt-auto rounded-lg border border-border bg-panel-2 px-3 py-2 text-xs text-slate-300">
       <div className="flex items-center gap-2">

ui/src/api.ts

@@ -108,6 +108,20 @@ export interface ChatErrorEvent {
   message: string;
 }
 
+export interface EngineDeployEvent {
+  model_id: string;
+  file_name: string;
+  state: string;
+  message: string | null;
+}
+
+export interface EngineDeployProgressEvent {
+  model_id: string;
+  file_name: string;
+  percent: number;
+  stage: string;
+}
+
 export interface DownloadProgressEvent {
   id: string;
   url: string;
@@ -148,6 +162,8 @@ export const api = {
     invoke<Message[]>("get_messages", { conversationId }),
   engineStart: (modelId: string, params: ChatParams) =>
     invoke<EngineStatus>("engine_start", { modelId, params }),
+  deployModel: (modelId: string, params: ChatParams) =>
+    invoke<void>("deploy_model", { modelId, params }),
   engineStop: () => invoke<void>("engine_stop"),
   engineStatus: () => invoke<EngineStatus>("engine_status"),
   chatSend: (payload: ChatSendPayload) => invoke<string>("chat_send", { payload }),

ui/src/components/Icon.tsx

@@ -0,0 +1,59 @@
+import type { ReactNode } from "react";
+
+const paths: Record<string, ReactNode> = {
+  chat: <path d="M21 15a2 2 0 0 1-2 2H7l-4 4V5a2 2 0 0 1 2-2h14a2 2 0 0 1 2 2z" />,
+  box: (
+    <>
+      <path d="M21 16V8a2 2 0 0 0-1-1.73l-7-4a2 2 0 0 0-2 0l-7 4A2 2 0 0 0 3 8v8a2 2 0 0 0 1 1.73l7 4a2 2 0 0 0 2 0l7-4A2 2 0 0 0 21 16z" />
+      <path d="M3.27 6.96L12 12.01l8.73-5.05" />
+      <path d="M12 22.08V12" />
+    </>
+  ),
+  plaza: (
+    <>
+      <rect x="3" y="3" width="7" height="7" rx="1" />
+      <rect x="14" y="3" width="7" height="7" rx="1" />
+      <rect x="3" y="14" width="7" height="7" rx="1" />
+      <rect x="14" y="14" width="7" height="7" rx="1" />
+    </>
+  ),
+  server: (
+    <>
+      <rect x="2" y="2" width="20" height="8" rx="2" />
+      <rect x="2" y="14" width="20" height="8" rx="2" />
+      <path d="M6 6h.01M6 18h.01" />
+    </>
+  ),
+  settings: (
+    <>
+      <path d="M4 21v-7M4 10V3M12 21v-9M12 8V3M20 21v-5M20 12V3" />
+      <path d="M1 14h6M9 8h6M17 16h6" />
+    </>
+  ),
+  "chevron-left": <path d="M15 18l-6-6 6-6" />,
+  "chevron-right": <path d="M9 18l6-6-6-6" />,
+};
+
+export default function Icon({
+  name,
+  className = "h-4 w-4",
+}: {
+  name: string;
+  className?: string;
+}) {
+  return (
+    <svg
+      viewBox="0 0 24 24"
+      fill="none"
+      stroke="currentColor"
+      strokeWidth="2"
+      strokeLinecap="round"
+      strokeLinejoin="round"
+      className={className}
+      aria-hidden="true"
+    >
+      {paths[name]}
+    </svg>
+  );
+}
+

ui/src/pages/ChatPage.tsx

@@ -10,6 +10,7 @@ import {
   onEvent,
 } from "../api";
 import { useStore } from "../store";
+import Icon from "../components/Icon";
 
 interface LocalMessage {
   id: string;
@@ -38,8 +39,19 @@ export default function ChatPage() {
   const [error, setError] = useState<string | null>(null);
   const [params, setParams] = useState<ChatParams>(defaultParams);
   const [selectedModelId, setSelectedModelId] = useState<string>("");
+  const [rightOpen, setRightOpen] = useState(
+    () => localStorage.getItem("xianren-right-open") !== "0",
+  );
   const bottomRef = useRef<HTMLDivElement>(null);
 
+  function toggleRight() {
+    setRightOpen((v) => {
+      const next = !v;
+      localStorage.setItem("xianren-right-open", next ? "1" : "0");
+      return next;
+    });
+  }
+
   useEffect(() => {
     if (models.length > 0 && !selectedModelId) {
       setSelectedModelId(models[0].id);
@@ -168,6 +180,9 @@ export default function ChatPage() {
   }
 
   const engineRunning = useStore((s) => s.engine?.running ?? false);
+  const engineModel = useStore((s) => s.engine?.model ?? null);
+  const deployStates = useStore((s) => s.deployStates);
+  const deployProgress = useStore((s) => s.deployProgress);
   const selectedIsRemote = useMemo(
     () => models.find((m) => m.id === selectedModelId)?.kind === "remote",
     [models, selectedModelId],
@@ -206,45 +221,6 @@ export default function ChatPage() {
             </div>
           ))}
         </div>
-        <div className="mt-3 border-t border-border pt-3">
-          <label className="mb-1 block text-xs text-slate-400">当前模型</label>
-          <select
-            className="input w-full"
-            value={selectedModelId}
-            onChange={(e) => setSelectedModelId(e.target.value)}
-          >
-            {models.map((m) => (
-              <option key={m.id} value={m.id}>
-                {m.file_name}
-                {m.kind === "remote" ? "（API）" : ""}
-              </option>
-            ))}
-          </select>
-          {selectedIsRemote ? (
-            <div className="mt-2 rounded-lg border border-sky-500/30 bg-sky-500/10 px-3 py-1.5 text-xs text-sky-300">
-              在线 API 模型，无需本地引擎
-            </div>
-          ) : (
-            <div className="mt-2 flex gap-2">
-              <button
-                className="flex-1 rounded-lg border border-border bg-panel-2 py-1.5 text-xs hover:bg-panel-2/60"
-                onClick={() =>
-                  api.engineStart(selectedModelId, params).catch((e) => alert(String(e)))
-                }
-                disabled={engineRunning}
-              >
-                启动引擎
-              </button>
-              <button
-                className="flex-1 rounded-lg border border-border bg-panel-2 py-1.5 text-xs hover:bg-panel-2/60"
-                onClick={() => api.engineStop().catch((e) => alert(String(e)))}
-                disabled={!engineRunning}
-              >
-                停止引擎
-              </button>
-            </div>
-          )}
-        </div>
       </div>
 
       <div className="flex flex-1 flex-col">
@@ -289,7 +265,121 @@ export default function ChatPage() {
         </div>
 
         <div className="border-t border-border bg-panel p-4">
-          <div className="mb-2 flex flex-wrap items-center gap-x-4 gap-y-1 text-xs text-slate-400">
+          <div className="flex items-end gap-3">
+            <textarea
+              className="input min-h-[72px] flex-1 resize-none"
+              placeholder={`向本地模型提问…（当前 ${engineRunning ? "引擎运行中" : "引擎未启动，发送时将自动启动"}）`}
+              value={input}
+              onChange={(e) => setInput(e.target.value)}
+              onKeyDown={(e) => {
+                if (e.key === "Enter" && !e.shiftKey) {
+                  e.preventDefault();
+                  handleSend();
+                }
+              }}
+            />
+            <button
+              className="btn-primary h-[72px] px-6"
+              onClick={handleSend}
+              disabled={streaming || !input.trim()}
+            >
+              {streaming ? "生成中…" : "发送"}
+            </button>
+          </div>
+        </div>
+      </div>
+
+      {/* 右侧边栏：当前模型 + 部署/请求参数（可收起） */}
+      {rightOpen ? (
+        <div className="flex w-72 flex-col overflow-y-auto border-l border-border bg-panel p-3">
+          <div className="mb-2 flex items-center justify-between">
+            <span className="text-xs uppercase text-slate-400">参数面板</span>
+            <button
+              className="rounded p-1 text-slate-400 hover:bg-panel-2/60 hover:text-slate-200"
+              onClick={toggleRight}
+              title="收起参数面板"
+            >
+              <Icon name="chevron-right" />
+            </button>
+          </div>
+          <div className="mb-1 text-xs text-slate-400">当前模型</div>
+          <select
+            className="input w-full"
+            value={selectedModelId}
+            onChange={(e) => setSelectedModelId(e.target.value)}
+          >
+            {models.map((m) => (
+              <option key={m.id} value={m.id}>
+                {m.file_name}
+                {m.kind === "remote" ? "（API）" : ""}
+                {engineRunning && engineModel === m.file_name ? " ●已部署" : ""}
+              </option>
+            ))}
+          </select>
+          {selectedIsRemote ? (
+            <div className="mt-2 rounded-lg border border-sky-500/30 bg-sky-500/10 px-3 py-1.5 text-xs text-sky-300">
+              在线 API 模型，无需本地引擎
+            </div>
+          ) : (
+            <div className="mt-2 flex gap-2">
+              <button
+                className="flex-1 rounded-lg border border-border bg-panel-2 py-1.5 text-xs hover:bg-panel-2/60"
+                onClick={() =>
+                  api.engineStart(selectedModelId, params).catch((e) => alert(String(e)))
+                }
+                disabled={engineRunning}
+              >
+                启动引擎
+              </button>
+              <button
+                className="flex-1 rounded-lg border border-border bg-panel-2 py-1.5 text-xs hover:bg-panel-2/60"
+                onClick={() => api.engineStop().catch((e) => alert(String(e)))}
+                disabled={!engineRunning}
+              >
+                停止引擎
+              </button>
+            </div>
+          )}
+          {deployStates[selectedModelId] === "loading" ? (
+            <div className="mt-2">
+              <div className="h-1.5 overflow-hidden rounded-full bg-panel-2">
+                <div
+                  className="h-full rounded-full bg-gradient-to-r from-accent to-accent-2 transition-all duration-300"
+                  style={{
+                    width: `${deployProgress[selectedModelId]?.percent ?? 8}%`,
+                  }}
+                />
+              </div>
+              <div className="mt-1 text-[10px] text-slate-400">
+                {deployProgress[selectedModelId]?.stage ?? "启动引擎进程"}{" "}
+                {deployProgress[selectedModelId]?.percent ?? 8}%
+              </div>
+            </div>
+          ) : null}
+          <div className="mt-1 truncate text-xs text-slate-500">{engineLabel}</div>
+
+          <div className="mt-4 border-t border-border pt-3">
+            <div className="mb-3 text-xs uppercase text-slate-400">部署参数</div>
+            <ParamSlider
+              label="上下文长度 ctx_size"
+              value={params.ctx_size}
+              min={512}
+              max={32768}
+              step={512}
+              onChange={(v) => setParams({ ...params, ctx_size: v })}
+            />
+            <ParamSlider
+              label="GPU 层数 ngl（99=全部）"
+              value={params.ngl}
+              min={0}
+              max={99}
+              step={1}
+              onChange={(v) => setParams({ ...params, ngl: v })}
+            />
+          </div>
+
+          <div className="mt-4 border-t border-border pt-3">
+            <div className="mb-3 text-xs uppercase text-slate-400">请求参数</div>
             <ParamSlider
               label="temperature"
               value={params.temperature}
@@ -314,47 +404,22 @@ export default function ChatPage() {
               step={64}
               onChange={(v) => setParams({ ...params, max_tokens: v })}
             />
-            <ParamSlider
-              label="ctx_size"
-              value={params.ctx_size}
-              min={512}
-              max={32768}
-              step={512}
-              onChange={(v) => setParams({ ...params, ctx_size: v })}
-            />
-            <ParamSlider
-              label="ngl(全部=99)"
-              value={params.ngl}
-              min={0}
-              max={99}
-              step={1}
-              onChange={(v) => setParams({ ...params, ngl: v })}
-            />
-            <span className="ml-auto text-slate-500">{engineLabel}</span>
-          </div>
-          <div className="flex items-end gap-3">
-            <textarea
-              className="input min-h-[72px] flex-1 resize-none"
-              placeholder={`向本地模型提问…（当前 ${engineRunning ? "引擎运行中" : "引擎未启动，发送时将自动启动"}）`}
-              value={input}
-              onChange={(e) => setInput(e.target.value)}
-              onKeyDown={(e) => {
-                if (e.key === "Enter" && !e.shiftKey) {
-                  e.preventDefault();
-                  handleSend();
-                }
-              }}
-            />
-            <button
-              className="btn-primary h-[72px] px-6"
-              onClick={handleSend}
-              disabled={streaming || !input.trim()}
-            >
-              {streaming ? "生成中…" : "发送"}
-            </button>
           </div>
         </div>
-      </div>
+      ) : (
+        <div className="flex w-9 flex-col items-center border-l border-border bg-panel py-2">
+          <button
+            className="rounded p-1.5 text-slate-400 hover:bg-panel-2/60 hover:text-slate-200"
+            onClick={toggleRight}
+            title="展开参数面板"
+          >
+            <Icon name="chevron-left" />
+          </button>
+          <span className="mt-3 whitespace-nowrap text-[10px] text-slate-500 [writing-mode:vertical-rl]">
+            参数
+          </span>
+        </div>
+      )}
     </div>
   );
 }
@@ -375,18 +440,20 @@ function ParamSlider({
   onChange: (v: number) => void;
 }) {
   return (
-    <label className="flex items-center gap-2">
-      <span className="w-24">{label}</span>
+    <label className="mb-3 block">
+      <div className="mb-1 flex items-center justify-between text-xs text-slate-400">
+        <span>{label}</span>
+        <span className="font-mono text-slate-300">{value}</span>
+      </div>
       <input
         type="range"
-        className="w-28 accent-indigo-500"
+        className="w-full accent-indigo-500"
         min={min}
         max={max}
         step={step}
         value={value}
         onChange={(e) => onChange(Number(e.target.value))}
       />
-      <span className="w-12 font-mono text-slate-300">{value}</span>
     </label>
   );
 }

ui/src/pages/ModelsPage.tsx

@@ -11,6 +11,9 @@ function fmtSize(bytes: number) {
 export default function ModelsPage() {
   const models = useStore((s) => s.models);
   const refreshModels = useStore((s) => s.refreshModels);
+  const engine = useStore((s) => s.engine);
+  const deployStates = useStore((s) => s.deployStates);
+  const deployProgress = useStore((s) => s.deployProgress);
   const [showRemoteForm, setShowRemoteForm] = useState(false);
   const [scanning, setScanning] = useState(false);
   const [remote, setRemote] = useState({
@@ -82,13 +85,36 @@ export default function ModelsPage() {
     await api.openPath(info.models_dir);
   }
 
+  async function handleDeploy(id: string, params?: Partial<import("../api").ChatParams>) {
+    try {
+      await api.deployModel(id, {
+        temperature: 0.7,
+        top_p: 0.9,
+        max_tokens: 2048,
+        ctx_size: 4096,
+        ngl: 99,
+        ...params,
+      });
+    } catch (e) {
+      alert(String(e));
+    }
+  }
+
+  async function handleStopEngine() {
+    try {
+      await api.engineStop();
+    } catch (e) {
+      alert(String(e));
+    }
+  }
+
   return (
     <div className="h-full overflow-auto p-6">
       <div className="mb-4 flex items-center justify-between">
         <div>
-          <h1 className="text-xl font-semibold">模型库</h1>
+          <h1 className="text-xl font-semibold">模型管理</h1>
           <p className="mt-1 text-sm text-slate-400">
-            启动时自动扫描模型目录，也可手动扫描；支持添加本地 GGUF 与在线 API 模型
+            启动时自动扫描模型目录，也可手动扫描；本地模型可点击「部署」后台加载
           </p>
         </div>
         <div className="flex gap-2">
@@ -208,19 +234,34 @@ export default function ModelsPage() {
                     {m.kind === "remote" ? "-" : fmtSize(m.file_size)}
                   </td>
                   <td className="px-4 py-3">
-                    {m.status === "ready" ? (
+                    {m.kind === "remote" ? (
+                      <span className="text-slate-400">-</span>
+                    ) : m.status === "ready" ? (
                       <span className="text-emerald-400">{m.status}</span>
                     ) : (
                       <span className="text-amber-400">{m.status}</span>
                     )}
                   </td>
-                  <td className="px-4 py-3 text-right">
-                    <button
-                      className="text-xs text-slate-400 hover:text-red-400"
-                      onClick={() => handleRemove(m.id, m.file_name)}
-                    >
-                      移除
-                    </button>
+                  <td className="px-4 py-3">
+                    <div className="flex items-center justify-end gap-2">
+                      {m.kind === "local" ? (
+                        <DeployButton
+                          model={m}
+                          engineRunning={engine?.running ?? false}
+                          deployed={engine?.model === m.file_name}
+                          deployState={deployStates[m.id]}
+                          deployProgress={deployProgress[m.id]}
+                          onDeploy={handleDeploy}
+                          onStop={handleStopEngine}
+                        />
+                      ) : null}
+                      <button
+                        className="text-xs text-slate-400 hover:text-red-400"
+                        onClick={() => handleRemove(m.id, m.file_name)}
+                      >
+                        移除
+                      </button>
+                    </div>
                   </td>
                 </tr>
               ))}
@@ -232,3 +273,61 @@ export default function ModelsPage() {
   );
 }
 
+function DeployButton({
+  model,
+  engineRunning,
+  deployed,
+  deployState,
+  deployProgress,
+  onDeploy,
+  onStop,
+}: {
+  model: { id: string; file_name: string };
+  engineRunning: boolean;
+  deployed: boolean;
+  deployState?: string;
+  deployProgress?: { percent: number; stage: string };
+  onDeploy: (id: string) => void;
+  onStop: () => void;
+}) {
+  if (deployed) {
+    return (
+      <div className="flex items-center gap-2">
+        <span className="rounded bg-emerald-500/20 px-2 py-1 text-xs text-emerald-300">
+          ● 已部署
+        </span>
+        <button
+          className="text-xs text-slate-400 hover:text-red-400"
+          onClick={onStop}
+          disabled={!engineRunning}
+        >
+          停止
+        </button>
+      </div>
+    );
+  }
+  if (deployState === "loading") {
+    const p = deployProgress ?? { percent: 8, stage: "启动引擎进程" };
+    return (
+      <div className="w-36">
+        <div className="h-1.5 overflow-hidden rounded-full bg-panel-2">
+          <div
+            className="h-full rounded-full bg-gradient-to-r from-accent to-accent-2 transition-all duration-300"
+            style={{ width: `${p.percent}%` }}
+          />
+        </div>
+        <div className="mt-1 truncate text-[10px] text-slate-400">
+          {p.stage} {p.percent}%
+        </div>
+      </div>
+    );
+  }
+  return (
+    <button
+      className="btn-secondary !px-3 !py-1 text-xs"
+      onClick={() => onDeploy(model.id)}
+    >
+      {deployState === "error" ? "重试部署" : "部署"}
+    </button>
+  );
+}

ui/src/pages/ServerPage.tsx

@@ -30,7 +30,7 @@ export default function ServerPage() {
 
   return (
     <div className="h-full overflow-auto p-6">
-      <h1 className="text-xl font-semibold">本地 API 服务</h1>
+      <h1 className="text-xl font-semibold">服务管理</h1>
       <p className="mt-1 text-sm text-slate-400">
         启动后即可用 OpenAI SDK / 任意 HTTP 客户端调用本地模型
       </p>
@@ -91,4 +91,3 @@ export default function ServerPage() {
     </div>
   );
 }
-

ui/src/store.ts

@@ -6,6 +6,10 @@ interface Store {
   engine: EngineStatus | null;
   server: ServerStatus | null;
   conversations: Conversation[];
+  deployStates: Record<string, string>;
+  deployProgress: Record<string, { percent: number; stage: string }>;
+  setDeployState: (modelId: string, state: string) => void;
+  setDeployProgress: (modelId: string, progress: { percent: number; stage: string }) => void;
   refreshModels: () => Promise<void>;
   refreshEngine: () => Promise<void>;
   refreshServer: () => Promise<void>;
@@ -17,10 +21,15 @@ export const useStore = create<Store>((set) => ({
   engine: null,
   server: null,
   conversations: [],
+  deployStates: {},
+  deployProgress: {},
+  setDeployState: (modelId, state) =>
+    set((s) => ({ deployStates: { ...s.deployStates, [modelId]: state } })),
+  setDeployProgress: (modelId, progress) =>
+    set((s) => ({ deployProgress: { ...s.deployProgress, [modelId]: progress } })),
   refreshModels: async () => set({ models: await api.listModels() }),
   refreshEngine: async () => set({ engine: await api.engineStatus() }),
   refreshServer: async () => set({ server: await api.serverStatus() }),
   refreshConversations: async () =>
     set({ conversations: await api.listConversations() }),
 }));
-