feat: hwaccel + Fanart/TheTVDB + AI + Discover + NFO + Tasks + Recycle bin

Backend - service/transcoder.go: encoder selector (software / nvenc / qsv / vaapi) with proper hwaccel + scale_* filters per encoder; configurable bitrate/preset/height/segment-seconds via transcoder.* config; new Active() snapshot + ActiveJob struct for the Tasks panel; unit-tested via buildFFmpegArgs(...). - service/thetvdb.go: TheTVDB v4 provider — login() caches the JWT for 24h, SearchSeries() returns Match struct. - service/fanart.go: Fanart.tv provider — high-res movie artwork keyed by TMDb id; used to upgrade poster/backdrop after a successful match. - service/scraper.go: provider chain reworked — anime → Bangumi, tv → TheTVDB, default → TMDb, with optional Fanart upgrade post-match. - service/discover.go: TMDb /trending/movie/day + /movie/popular for the Discover rail. - service/ai.go: OpenAI-compatible client. SmartSearch() turns a free- form query into a structured SearchIntent JSON; Recommend() emits a short list of titles given the user's history. Disabled when ai.api_key is empty. - service/nfo.go: per-movie .nfo writer (Kodi/Jellyfin schema) + library-scope batch exporter. - service/media.go: SoftDelete / Restore / Purge / ListRecycleBin helpers backed by gorm Unscoped(). - service/service.go: container wires Fanart, TheTVDB, Discover, AI, NFO; everything still tears down cleanly via Close(). - config: new transcoder.* section (encoder, preset, video_bitrate, max_rate, buf_size, max_height, segment_seconds) with sensible defaults. Handlers / routes - new files: discover.go, recycle.go, nfo.go, ai.go, tasks.go. - registers GET /tasks, /discover/{trending,popular}, /ai/{status,search,recommend}; admin-only GET /recycle and DELETE/POST /media/:id (soft delete / restore / purge); admin-only POST /media/:id/nfo and /libraries/:id/nfo. Frontend - new pages: DiscoverPage, TasksPage, RecycleBinPage; SearchPage gains an optional AI smart-search toggle that calls /api/ai/search. - MediaDetailPage gains admin actions for 导出 NFO and 移至回收站. - new api/ helpers: ai.ts, discover.ts, recycle.ts, tasks.ts. - App.tsx routes /discover, /tasks, /recycle (last two admin-only). - Layout sidebar groups now include Discover + Tasks + Recycle Bin. Docker / config - Dockerfile: adds intel-media-driver / libva-utils / mesa-va-gallium so QSV + VAAPI work out of the box on Intel iGPUs. - docker-compose.yml: documents devices + group_add + nvidia runtime overrides for hardware transcoding. Verified: go build, go vet, go test (incl. new TestBuildFFmpegArgs across software / nvenc / qsv / vaapi encoder profiles); tsc -b && vite build emits 22 route chunks plus the deferred hls chunk; main bundle 248 KB / 83 KB gzipped.
2026-09-28 03:06:38 +08:00 · 2026-05-14 16:22:08 +00:00
parent 7bd6bcfb1b
commit 8824638efb
32 changed files with 1795 additions and 83 deletions
@@ -33,8 +33,14 @@ RUN CGO_ENABLED=0 GOOS=${TARGETOS} GOARCH=${TARGETARCH} \

 # ---- Stage 3: runtime ------------------------------------------------------
 FROM alpine:3.19
+# ffmpeg/intel-media-driver/libva enable QSV / VAAPI on Intel iGPUs.
+# NVENC requires the proprietary NVIDIA Container Toolkit on the host —
+# no extra packages needed inside the image, only --gpus all on docker run.
 RUN apk add --no-cache \
        ffmpeg \
+        intel-media-driver \
+        libva-utils \
+        mesa-va-gallium \
        tzdata \
        ca-certificates \
        wget \
@@ -42,7 +42,10 @@ making deployment painless on NAS hardware.
 - ✅ Library CRUD + recursive filesystem scan
 - ✅ ffprobe metadata extraction (duration / resolution / codecs / container)
 - ✅ Scene-noise filename cleaner with year + season/episode parsing
- ✅ TMDb scraper (movies) + Bangumi scraper (anime), provider chain by library type
+- ✅ Multi-provider scrape chain by library type:
+  - movie → TMDb (with optional Fanart.tv high-res poster upgrade)
+  - tv → TheTVDB (fallback TMDb)
+  - anime → Bangumi (fallback TMDb)
 - ✅ Image proxy with disk cache for TMDb / Bangumi / Douban / Fanart / TheTVDB
 - ✅ TV / anime libraries grouped by season with episode listing
 - ✅ fsnotify-based filesystem watcher with 5 s coalescing debouncer
@@ -61,8 +64,17 @@ making deployment painless on NAS hardware.
 ### Operations
 - ✅ Real-time scan / scrape / transcode / download / subscription events over WebSocket
 - ✅ Operator dashboard at `/stats` (CPU / memory / disk / library counts / Goroutines)
+- ✅ Real-time tasks panel at `/tasks` (active ffmpeg jobs + qBittorrent torrents)
+- ✅ Recycle bin at `/recycle` (soft delete + restore + purge)
+- ✅ NFO export (Kodi / Jellyfin compatibility) — single media or whole library
+- ✅ Hardware-accel encoder profiles: software / NVENC / Intel QSV / VAAPI
 - ✅ Single-binary build, multi-arch Docker image, GitHub Actions CI + GHCR publish

+### Discovery & AI
+- ✅ TMDb Discover — Trending (today) + Popular rails on `/discover`
+- ✅ AI smart search (OpenAI-compatible) — natural-language queries → structured intent
+- ✅ AI recommendations seeded from your watch history (`GET /api/ai/recommend`)
+
 ### Frontend
 - ✅ React SPA with code-splitting: Login / Home / Library / Search / Favourites /
  Playlists / Media detail / Player (HLS + direct + subtitles) / Profile /
@@ -74,12 +86,10 @@ making deployment painless on NAS hardware.

 | Area | Status |
 |------|--------|
-| Hardware-accelerated transcoding (NVENC / QSV / VAAPI) profiles | ⏳ |
-| Douban / Fanart / TheTVDB providers | ⏳ |
-| Subtitle search (online providers) | ⏳ |
-| Emby/Jellyfin compatibility layer | ⏳ |
+| Bidirectional Jellyfin / Emby compatibility layer | ⏳ |
 | DLNA / Chromecast | ⏳ |
-| AI metadata enhancement & smart search | ⏳ |
+| Online subtitle search providers | ⏳ |
+| Multi-bitrate ABR transcode profiles | ⏳ |

 ---

@@ -64,6 +64,20 @@ media:
  tv_dir: ""
  anime_dir: ""

+transcoder:
+  # Hardware encoder for HLS transcodes.
+  #   ""        software libx264 (default; runs anywhere)
+  #   nvenc     h264_nvenc      (NVIDIA GPU, requires --gpus all)
+  #   qsv       h264_qsv        (Intel iGPU, requires /dev/dri)
+  #   vaapi     h264_vaapi      (Intel/AMD VAAPI, requires /dev/dri)
+  encoder: ""
+  preset: veryfast
+  video_bitrate: 1500k
+  max_rate: 1800k
+  buf_size: 3000k
+  max_height: 720
+  segment_seconds: 4
+
 ai:
  enabled: false
  provider: openai           # openai | deepseek | qwen | ollama
@@ -34,15 +34,27 @@ services:
      MEDIASTATION_DATABASE_DB_PATH: /data/mediastation.db
      MEDIASTATION_CACHE_CACHE_DIR: /cache
      MEDIASTATION_LOGGING_LEVEL: info
+      # MEDIASTATION_TRANSCODER_ENCODER: vaapi    # qsv | vaapi | nvenc | (empty=software)
+      # MEDIASTATION_TRANSCODER_MAX_HEIGHT: 1080
      # MEDIASTATION_SECRETS_JWT_SECRET: change-me-to-a-strong-random-string
      # MEDIASTATION_APP_CORS_ORIGINS: '["https://media.example.com"]'
      PUID: "1000"
      PGID: "1000"
      TZ: Asia/Shanghai

-    # Hardware-accelerated transcoding (uncomment when you have an Intel iGPU).
+    # Hardware-accelerated transcoding.
+    #   * Intel iGPU (QSV / VAAPI): mount /dev/dri.
+    #   * NVIDIA NVENC: switch the runtime to nvidia and add `gpus: all`.
    # devices:
    #   - /dev/dri:/dev/dri
+    # group_add:
+    #   - "989"   # match render GID on host
+    # runtime: nvidia
+    # deploy:
+    #   resources:
+    #     reservations:
+    #       devices:
+    #         - capabilities: [gpu]

    healthcheck:
      test: ["CMD-SHELL", "wget -qO- http://127.0.0.1:8080/api/health || exit 1"]
@@ -30,13 +30,25 @@ const EnvPrefix = "MEDIASTATION"

 // Config is the root config aggregate.
 type Config struct {
-	App      AppConfig      `mapstructure:"app"`
-	Database DatabaseConfig `mapstructure:"database"`
-	Secrets  SecretsConfig  `mapstructure:"secrets"`
-	Logging  LoggingConfig  `mapstructure:"logging"`
-	Cache    CacheConfig    `mapstructure:"cache"`
-	Media    MediaConfig    `mapstructure:"media"`
-	AI       AIConfig       `mapstructure:"ai"`
+	App        AppConfig        `mapstructure:"app"`
+	Database   DatabaseConfig   `mapstructure:"database"`
+	Secrets    SecretsConfig    `mapstructure:"secrets"`
+	Logging    LoggingConfig    `mapstructure:"logging"`
+	Cache      CacheConfig      `mapstructure:"cache"`
+	Media      MediaConfig      `mapstructure:"media"`
+	Transcoder TranscoderConfig `mapstructure:"transcoder"`
+	AI         AIConfig         `mapstructure:"ai"`
+}
+
+// TranscoderConfig controls the HLS / ffmpeg backend.
+type TranscoderConfig struct {
+	Encoder        string `mapstructure:"encoder"` // "" / nvenc / qsv / vaapi
+	Preset         string `mapstructure:"preset"`
+	VideoBitrate   string `mapstructure:"video_bitrate"`
+	MaxRate        string `mapstructure:"max_rate"`
+	BufSize        string `mapstructure:"buf_size"`
+	MaxHeight      int    `mapstructure:"max_height"`
+	SegmentSeconds int    `mapstructure:"segment_seconds"`
 }

 // AppConfig holds runtime app parameters.
@@ -195,6 +207,14 @@ func setDefaults(v *viper.Viper) {
 	v.SetDefault("ai.model", "gpt-4o-mini")
 	v.SetDefault("ai.timeout", 30)
 	v.SetDefault("ai.max_concurrent", 3)
+
+	v.SetDefault("transcoder.encoder", "")
+	v.SetDefault("transcoder.preset", "veryfast")
+	v.SetDefault("transcoder.video_bitrate", "1500k")
+	v.SetDefault("transcoder.max_rate", "1800k")
+	v.SetDefault("transcoder.buf_size", "3000k")
+	v.SetDefault("transcoder.max_height", 720)
+	v.SetDefault("transcoder.segment_seconds", 4)
 }

 // normalize fills derived defaults and self-heals empty critical fields.
@@ -0,0 +1,71 @@
+// Package handler — AI integration endpoints.
+package handler
+
+import (
+	"net/http"
+	"strings"
+
+	"github.com/gin-gonic/gin"
+
+	"github.com/ShukeBta/MediaStationGo/internal/middleware"
+	"github.com/ShukeBta/MediaStationGo/internal/service"
+)
+
+type smartSearchReq struct {
+	Query string `json:"query" binding:"required"`
+}
+
+func smartSearchHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		var req smartSearchReq
+		if err := c.ShouldBindJSON(&req); err != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
+			return
+		}
+		intent, err := svc.AI.SmartSearch(c.Request.Context(), req.Query)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		// Run the actual library search using the cleaned query so the
+		// caller can render results in one round-trip.
+		items, _ := svc.Media.SearchMedia(c.Request.Context(), intent.Query, 60)
+		c.JSON(http.StatusOK, gin.H{
+			"intent": intent,
+			"items":  items,
+		})
+	}
+}
+
+func aiRecommendHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		uid, _ := c.Get(middleware.CtxUserID)
+		hist, err := svc.Playback.RecentHistory(c.Request.Context(), toString(uid), 10)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		titles := make([]string, 0, len(hist))
+		for _, h := range hist {
+			if h.Media != nil && strings.TrimSpace(h.Media.Title) != "" {
+				titles = append(titles, h.Media.Title)
+			}
+		}
+		out, err := svc.AI.Recommend(c.Request.Context(), titles, 8)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.JSON(http.StatusOK, gin.H{"titles": out})
+	}
+}
+
+func aiStatusHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		c.JSON(http.StatusOK, gin.H{
+			"enabled":  svc.AI.Enabled(),
+			"provider": svc.Cfg.AI.Provider,
+			"model":    svc.Cfg.AI.Model,
+		})
+	}
+}
@@ -0,0 +1,32 @@
+// Package handler — TMDb discovery endpoints.
+package handler
+
+import (
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+
+	"github.com/ShukeBta/MediaStationGo/internal/service"
+)
+
+func trendingHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		items, err := svc.Discover.Trending(c.Request.Context())
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.JSON(http.StatusOK, gin.H{"items": items})
+	}
+}
+
+func popularHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		items, err := svc.Discover.Popular(c.Request.Context())
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.JSON(http.StatusOK, gin.H{"items": items})
+	}
+}
@@ -50,8 +50,13 @@ func Register(r *gin.Engine, cfg *config.Config, log *zap.Logger, svc *service.C
 			authed.GET("/media", searchMediaHandler(svc))
 			authed.POST("/media/:id/scrape", middleware.AdminRequired(), scrapeOneHandler(svc))
 			authed.POST("/media/:id/probe", middleware.AdminRequired(), reprobeHandler(svc))
+			authed.DELETE("/media/:id", middleware.AdminRequired(), deleteMediaHandler(svc))
+			authed.POST("/media/:id/restore", middleware.AdminRequired(), restoreMediaHandler(svc))
+			authed.DELETE("/media/:id/purge", middleware.AdminRequired(), purgeMediaHandler(svc))
 			authed.GET("/media/:id/subtitles", listSubtitlesHandler(svc))
 			authed.GET("/subtitles/:id", serveSubtitleHandler(svc))
+			authed.POST("/media/:id/nfo", middleware.AdminRequired(), exportNFOHandler(svc))
+			authed.POST("/libraries/:id/nfo", middleware.AdminRequired(), exportLibraryNFOHandler(svc))

 			// Streaming.
 			authed.GET("/stream/:id", streamHandler(svc))
@@ -90,6 +95,19 @@ func Register(r *gin.Engine, cfg *config.Config, log *zap.Logger, svc *service.C

 			// Stats / dashboard.
 			authed.GET("/stats", statsHandler(svc))
+			authed.GET("/tasks", tasksHandler(svc))
+
+			// Discover (TMDb trending / popular).
+			authed.GET("/discover/trending", trendingHandler(svc))
+			authed.GET("/discover/popular", popularHandler(svc))
+
+			// AI.
+			authed.GET("/ai/status", aiStatusHandler(svc))
+			authed.POST("/ai/search", smartSearchHandler(svc))
+			authed.GET("/ai/recommend", aiRecommendHandler(svc))
+
+			// Recycle bin.
+			authed.GET("/recycle", middleware.AdminRequired(), listRecycleHandler(svc))

 			authed.GET("/ws", wsHandler(svc))
 		}
@@ -0,0 +1,32 @@
+// Package handler — NFO export endpoints (Kodi / Jellyfin compatibility).
+package handler
+
+import (
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+
+	"github.com/ShukeBta/MediaStationGo/internal/service"
+)
+
+func exportNFOHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		path, err := svc.NFO.ExportOne(c.Request.Context(), c.Param("id"))
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.JSON(http.StatusOK, gin.H{"path": path})
+	}
+}
+
+func exportLibraryNFOHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		written, err := svc.NFO.ExportLibrary(c.Request.Context(), c.Param("id"))
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.JSON(http.StatusOK, gin.H{"written": written})
+	}
+}
@@ -0,0 +1,51 @@
+// Package handler — recycle bin endpoints.
+package handler
+
+import (
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+
+	"github.com/ShukeBta/MediaStationGo/internal/service"
+)
+
+func deleteMediaHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		if err := svc.Media.SoftDelete(c.Request.Context(), c.Param("id")); err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.Status(http.StatusNoContent)
+	}
+}
+
+func listRecycleHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		items, err := svc.Media.ListRecycleBin(c.Request.Context(), 200)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.JSON(http.StatusOK, gin.H{"items": items})
+	}
+}
+
+func restoreMediaHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		if err := svc.Media.RestoreDeleted(c.Request.Context(), c.Param("id")); err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.Status(http.StatusNoContent)
+	}
+}
+
+func purgeMediaHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		if err := svc.Media.PurgeDeleted(c.Request.Context(), c.Param("id")); err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.Status(http.StatusNoContent)
+	}
+}
@@ -0,0 +1,26 @@
+// Package handler — live tasks board.
+//
+// /api/tasks aggregates running ffmpeg transcodes, qBittorrent torrents
+// and recent scrape progress into a single snapshot suitable for the
+// React Tasks panel. The panel can layer this REST snapshot on top of
+// live WS events for instant updates.
+package handler
+
+import (
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+
+	"github.com/ShukeBta/MediaStationGo/internal/service"
+)
+
+func tasksHandler(svc *service.Container) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		transcodes := svc.Transcoder.Active()
+		_, torrents, _ := svc.Downloads.List(c.Request.Context())
+		c.JSON(http.StatusOK, gin.H{
+			"transcodes": transcodes,
+			"torrents":   torrents,
+		})
+	}
+}
@@ -0,0 +1,166 @@
+// Package service — AI integration (OpenAI-compatible chat completions).
+//
+// AIService is a thin wrapper around any OpenAI-compatible REST endpoint
+// (OpenAI, DeepSeek, Qwen, Ollama, …). Today we expose two operations:
+//
+//   - SmartSearch:    interpret a free-form Chinese / English query and
+//                     return a normalised JSON intent the React UI can
+//                     translate into filter params.
+//   - Recommend:      given a list of recently-watched titles, generate
+//                     a short list of "you might like…" recommendations.
+//
+// The service is disabled (every method returns nil) when ai.enabled is
+// false or ai.api_key is empty.
+package service
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+
+	"go.uber.org/zap"
+
+	"github.com/ShukeBta/MediaStationGo/internal/config"
+)
+
+// AIService talks to an OpenAI-compatible chat-completions endpoint.
+type AIService struct {
+	cfg    *config.Config
+	log    *zap.Logger
+	client *http.Client
+}
+
+// NewAIService is the constructor.
+func NewAIService(cfg *config.Config, log *zap.Logger) *AIService {
+	timeout := time.Duration(cfg.AI.Timeout) * time.Second
+	if timeout <= 0 {
+		timeout = 30 * time.Second
+	}
+	return &AIService{
+		cfg:    cfg,
+		log:    log,
+		client: &http.Client{Timeout: timeout},
+	}
+}
+
+// Enabled reports whether the AI integration is configured.
+func (a *AIService) Enabled() bool {
+	return a.cfg.AI.Enabled && strings.TrimSpace(a.cfg.AI.APIKey) != ""
+}
+
+// SearchIntent is the structured output the smart search endpoint returns.
+type SearchIntent struct {
+	Query    string `json:"query"`
+	Year     int    `json:"year,omitempty"`
+	Genre    string `json:"genre,omitempty"`
+	Type     string `json:"type,omitempty"` // movie / tv / anime / music
+	Sort     string `json:"sort,omitempty"` // recent / rating / random
+	Language string `json:"language,omitempty"`
+}
+
+// SmartSearch turns a natural-language query into a structured intent.
+// Returns a best-effort intent on parse failure (raw query passes through).
+func (a *AIService) SmartSearch(ctx context.Context, raw string) (*SearchIntent, error) {
+	if !a.Enabled() {
+		return &SearchIntent{Query: raw}, nil
+	}
+	const sys = "You are a media-library search assistant. Read the user's query and " +
+		"output a JSON object with the keys: query (string), year (int, optional), " +
+		"genre (string, optional), type (movie|tv|anime|music, optional), sort " +
+		"(recent|rating|random, optional), language (zh|en, optional). Respond with " +
+		"JSON only, no commentary."
+	out, err := a.complete(ctx, sys, raw)
+	if err != nil {
+		return &SearchIntent{Query: raw}, err
+	}
+	var intent SearchIntent
+	if err := json.Unmarshal([]byte(out), &intent); err != nil {
+		// Fallback: tolerate non-JSON output by treating the raw text as
+		// the cleaned query.
+		intent.Query = strings.TrimSpace(out)
+	}
+	if intent.Query == "" {
+		intent.Query = raw
+	}
+	return &intent, nil
+}
+
+// Recommend builds a short comma-separated list of titles given the user's
+// history. The first call is intentionally best-effort: a future iteration
+// may chain media DB lookups onto each suggestion.
+func (a *AIService) Recommend(ctx context.Context, history []string, max int) ([]string, error) {
+	if !a.Enabled() || len(history) == 0 {
+		return nil, nil
+	}
+	if max <= 0 || max > 20 {
+		max = 8
+	}
+	sys := fmt.Sprintf("You are a film / TV recommendation assistant. Reply with %d "+
+		"comma-separated titles only, no commentary, in the same language as the input.", max)
+	usr := "I recently watched: " + strings.Join(history, "; ")
+	out, err := a.complete(ctx, sys, usr)
+	if err != nil {
+		return nil, err
+	}
+	parts := strings.Split(out, ",")
+	titles := make([]string, 0, len(parts))
+	for _, p := range parts {
+		p = strings.TrimSpace(p)
+		p = strings.Trim(p, "\"'`")
+		if p != "" {
+			titles = append(titles, p)
+		}
+	}
+	return titles, nil
+}
+
+// complete is the shared helper — POST /v1/chat/completions.
+func (a *AIService) complete(ctx context.Context, system, user string) (string, error) {
+	payload := map[string]any{
+		"model":       a.cfg.AI.Model,
+		"temperature": 0.2,
+		"messages": []map[string]string{
+			{"role": "system", "content": system},
+			{"role": "user", "content": user},
+		},
+	}
+	body, _ := json.Marshal(payload)
+	endpoint := strings.TrimRight(a.cfg.AI.APIBase, "/") + "/chat/completions"
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, endpoint, bytes.NewReader(body))
+	if err != nil {
+		return "", err
+	}
+	req.Header.Set("Content-Type", "application/json")
+	req.Header.Set("Authorization", "Bearer "+a.cfg.AI.APIKey)
+	resp, err := a.client.Do(req)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode >= 400 {
+		raw, _ := io.ReadAll(resp.Body)
+		return "", fmt.Errorf("ai %d: %s", resp.StatusCode, strings.TrimSpace(string(raw)))
+	}
+
+	type choice struct {
+		Message struct {
+			Content string `json:"content"`
+		} `json:"message"`
+	}
+	var out struct {
+		Choices []choice `json:"choices"`
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&out); err != nil {
+		return "", err
+	}
+	if len(out.Choices) == 0 {
+		return "", errors.New("ai: empty completion")
+	}
+	return strings.TrimSpace(out.Choices[0].Message.Content), nil
+}
@@ -0,0 +1,117 @@
+// Package service — TMDb discovery (trending / popular).
+//
+// DiscoverService surfaces curated lists from TMDb so the React home
+// page can show "Trending" and "Popular" rails alongside the user's own
+// library. All methods gracefully no-op when the TMDb provider is
+// disabled.
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/url"
+	"time"
+
+	"go.uber.org/zap"
+)
+
+// DiscoverService talks to TMDb's /trending and /movie/popular endpoints.
+type DiscoverService struct {
+	log    *zap.Logger
+	tmdb   *TMDbProvider
+	client *http.Client
+}
+
+// NewDiscoverService is the constructor.
+func NewDiscoverService(log *zap.Logger, tmdb *TMDbProvider) *DiscoverService {
+	return &DiscoverService{
+		log:    log,
+		tmdb:   tmdb,
+		client: &http.Client{Timeout: 15 * time.Second},
+	}
+}
+
+// Trending returns the daily trending movies (TMDb /trending/movie/day).
+func (d *DiscoverService) Trending(ctx context.Context) ([]Match, error) {
+	return d.fetch(ctx, "/trending/movie/day")
+}
+
+// Popular returns the popular movies list (TMDb /movie/popular).
+func (d *DiscoverService) Popular(ctx context.Context) ([]Match, error) {
+	return d.fetch(ctx, "/movie/popular")
+}
+
+// fetch is the shared helper that paginates page=1 only — that's all the
+// home page needs and it keeps us under TMDb's 50 rps limit.
+func (d *DiscoverService) fetch(ctx context.Context, path string) ([]Match, error) {
+	if d.tmdb == nil || !d.tmdb.Enabled() {
+		return nil, nil
+	}
+	q := url.Values{}
+	q.Set("api_key", d.tmdb.cfg.Secrets.TMDbAPIKey)
+	q.Set("language", "zh-CN")
+	q.Set("page", "1")
+	u := d.tmdb.base + path + "?" + q.Encode()
+
+	type result struct {
+		ID           int     `json:"id"`
+		Title        string  `json:"title"`
+		Name         string  `json:"name"`
+		Overview     string  `json:"overview"`
+		PosterPath   string  `json:"poster_path"`
+		BackdropPath string  `json:"backdrop_path"`
+		ReleaseDate  string  `json:"release_date"`
+		FirstAirDate string  `json:"first_air_date"`
+		VoteAverage  float32 `json:"vote_average"`
+	}
+	type page struct {
+		Results []result `json:"results"`
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, u, nil)
+	if err != nil {
+		return nil, err
+	}
+	resp, err := d.client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode >= 400 {
+		return nil, fmt.Errorf("tmdb %s: %d", path, resp.StatusCode)
+	}
+	var p page
+	if err := json.NewDecoder(resp.Body).Decode(&p); err != nil {
+		return nil, err
+	}
+	out := make([]Match, 0, len(p.Results))
+	for _, r := range p.Results {
+		title := r.Title
+		if title == "" {
+			title = r.Name
+		}
+		m := Match{
+			TMDbID:   r.ID,
+			Title:    title,
+			Overview: r.Overview,
+			Rating:   r.VoteAverage,
+		}
+		if r.PosterPath != "" {
+			m.PosterURL = d.tmdb.imgCDN + "/w500" + r.PosterPath
+		}
+		if r.BackdropPath != "" {
+			m.BackdropURL = d.tmdb.imgCDN + "/w1280" + r.BackdropPath
+		}
+		date := r.ReleaseDate
+		if date == "" {
+			date = r.FirstAirDate
+		}
+		if len(date) >= 4 {
+			fmt.Sscanf(date[:4], "%d", &m.Year)
+		}
+		out = append(out, m)
+	}
+	return out, nil
+}
@@ -0,0 +1,109 @@
+// Package service — Fanart.tv image provider.
+//
+// Fanart.tv (https://fanart.tv) hosts community-curated artwork. We use
+// it to upgrade movie / tv posters and backdrops with higher-resolution
+// alternatives once a TMDb / Bangumi match is established.
+//
+// Endpoints used:
+//
+//   GET /v3/movies/{tmdb_id}        (artwork keyed by TMDb id)
+//   GET /v3/tv/{thetvdb_id}         (artwork keyed by TheTVDB id)
+//
+// The provider is enabled iff secrets.fanart_tv_api_key is non-empty.
+package service
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"time"
+
+	"go.uber.org/zap"
+
+	"github.com/ShukeBta/MediaStationGo/internal/config"
+)
+
+// FanartProvider talks to https://webservice.fanart.tv.
+type FanartProvider struct {
+	cfg    *config.Config
+	log    *zap.Logger
+	client *http.Client
+}
+
+// NewFanartProvider is the constructor.
+func NewFanartProvider(cfg *config.Config, log *zap.Logger) *FanartProvider {
+	return &FanartProvider{
+		cfg:    cfg,
+		log:    log,
+		client: &http.Client{Timeout: 15 * time.Second},
+	}
+}
+
+// Enabled reports whether an API key is configured.
+func (f *FanartProvider) Enabled() bool { return f.cfg.Secrets.FanartAPIKey != "" }
+
+// Artwork is the high-res image set returned by Fanart.tv.
+type Artwork struct {
+	Poster   string `json:"poster"`
+	Backdrop string `json:"backdrop"`
+	Logo     string `json:"logo"`
+	Thumb    string `json:"thumb"`
+}
+
+// MovieArtwork looks up the artwork bundle for a TMDb movie id.
+func (f *FanartProvider) MovieArtwork(ctx context.Context, tmdbID int) (*Artwork, error) {
+	if !f.Enabled() || tmdbID <= 0 {
+		return nil, nil
+	}
+	type entry struct {
+		URL  string `json:"url"`
+		Lang string `json:"lang"`
+	}
+	type page struct {
+		MoviePoster   []entry `json:"movieposter"`
+		MovieBackgr   []entry `json:"moviebackground"`
+		HDLogo        []entry `json:"hdmovielogo"`
+		MovieThumb    []entry `json:"moviethumb"`
+	}
+	u := fmt.Sprintf("https://webservice.fanart.tv/v3/movies/%d?api_key=%s",
+		tmdbID, f.cfg.Secrets.FanartAPIKey)
+	var p page
+	if err := f.getJSON(ctx, u, &p); err != nil {
+		return nil, err
+	}
+	a := &Artwork{}
+	if len(p.MoviePoster) > 0 {
+		a.Poster = p.MoviePoster[0].URL
+	}
+	if len(p.MovieBackgr) > 0 {
+		a.Backdrop = p.MovieBackgr[0].URL
+	}
+	if len(p.HDLogo) > 0 {
+		a.Logo = p.HDLogo[0].URL
+	}
+	if len(p.MovieThumb) > 0 {
+		a.Thumb = p.MovieThumb[0].URL
+	}
+	return a, nil
+}
+
+func (f *FanartProvider) getJSON(ctx context.Context, u string, out any) error {
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, u, nil)
+	if err != nil {
+		return err
+	}
+	req.Header.Set("User-Agent", "MediaStationGo/0.1")
+	resp, err := f.client.Do(req)
+	if err != nil {
+		return err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode == http.StatusNotFound {
+		return nil
+	}
+	if resp.StatusCode >= 400 {
+		return fmt.Errorf("fanart %s: %d", u, resp.StatusCode)
+	}
+	return json.NewDecoder(resp.Body).Decode(out)
+}
@@ -86,3 +86,34 @@ func (s *MediaService) SearchMedia(ctx context.Context, query string, limit int)
 func (s *MediaService) GetMedia(ctx context.Context, id string) (*model.Media, error) {
 	return s.repo.Media.FindByID(ctx, id)
 }
+
+// SoftDelete moves a media row to the recycle bin (gorm soft delete).
+// The on-disk file is kept; admins can purge it later.
+func (s *MediaService) SoftDelete(ctx context.Context, id string) error {
+	return s.repo.DB.Where("id = ?", id).Delete(&model.Media{}).Error
+}
+
+// RestoreDeleted unsets DeletedAt for a single media row.
+func (s *MediaService) RestoreDeleted(ctx context.Context, id string) error {
+	return s.repo.DB.Unscoped().Model(&model.Media{}).
+		Where("id = ?", id).Update("deleted_at", nil).Error
+}
+
+// ListRecycleBin returns every soft-deleted row, newest first.
+func (s *MediaService) ListRecycleBin(ctx context.Context, limit int) ([]model.Media, error) {
+	if limit <= 0 || limit > 500 {
+		limit = 100
+	}
+	var rows []model.Media
+	err := s.repo.DB.Unscoped().
+		Where("deleted_at IS NOT NULL").
+		Order("deleted_at desc").
+		Limit(limit).
+		Find(&rows).Error
+	return rows, err
+}
+
+// PurgeDeleted permanently removes a soft-deleted row from the database.
+func (s *MediaService) PurgeDeleted(ctx context.Context, id string) error {
+	return s.repo.DB.Unscoped().Where("id = ?", id).Delete(&model.Media{}).Error
+}
@@ -0,0 +1,115 @@
+// Package service — NFO writer (Kodi / Jellyfin compatibility).
+//
+// Kodi and Jellyfin index media using sidecar XML files alongside the
+// source video. We export a minimal subset that those scrapers consume
+// happily:
+//
+//   movie.mkv    -> movie.nfo            (<movie>...</movie>)
+//   tvshow/      -> tvshow.nfo           (<tvshow>...</tvshow>)  [future]
+//
+// Today only the per-movie writer is implemented; the per-show / per-episode
+// writers are stubbed with TODO markers.
+package service
+
+import (
+	"context"
+	"encoding/xml"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"go.uber.org/zap"
+
+	"github.com/ShukeBta/MediaStationGo/internal/repository"
+)
+
+// NFOService is the entry point used by the admin "导出 NFO" action.
+type NFOService struct {
+	log  *zap.Logger
+	repo *repository.Container
+}
+
+// NewNFOService is the constructor.
+func NewNFOService(log *zap.Logger, repo *repository.Container) *NFOService {
+	return &NFOService{log: log, repo: repo}
+}
+
+// movieNFO is the on-disk schema. Tag names match Kodi's expectations.
+type movieNFO struct {
+	XMLName  xml.Name `xml:"movie"`
+	Title    string   `xml:"title"`
+	Original string   `xml:"originaltitle,omitempty"`
+	Year     int      `xml:"year,omitempty"`
+	Plot     string   `xml:"plot,omitempty"`
+	Rating   float32  `xml:"rating,omitempty"`
+	Poster   string   `xml:"thumb,omitempty"`
+	Fanart   string   `xml:"fanart,omitempty"`
+	TMDb     int      `xml:"tmdbid,omitempty"`
+}
+
+// ExportOne writes a movie.nfo file next to the media file. Existing files
+// are overwritten so a re-scrape always reflects the latest metadata.
+func (s *NFOService) ExportOne(ctx context.Context, mediaID string) (string, error) {
+	m, err := s.repo.Media.FindByID(ctx, mediaID)
+	if err != nil {
+		return "", err
+	}
+	if m == nil {
+		return "", errors.New("media not found")
+	}
+	if m.Path == "" {
+		return "", errors.New("media has empty path")
+	}
+
+	doc := movieNFO{
+		Title:    m.Title,
+		Original: m.OriginalName,
+		Year:     m.Year,
+		Plot:     m.Overview,
+		Rating:   m.Rating,
+		Poster:   m.PosterURL,
+		Fanart:   m.BackdropURL,
+		TMDb:     m.TMDbID,
+	}
+	out, err := xml.MarshalIndent(doc, "", "  ")
+	if err != nil {
+		return "", err
+	}
+	body := []byte(xml.Header + string(out) + "\n")
+
+	dst := nfoPath(m.Path)
+	if err := os.WriteFile(dst, body, 0o644); err != nil {
+		return "", err
+	}
+	s.log.Info("nfo exported", zap.String("media_id", m.ID), zap.String("path", dst))
+	return dst, nil
+}
+
+// ExportLibrary loops through every matched movie in a library and writes
+// an .nfo for each. Returns (written, error).
+func (s *NFOService) ExportLibrary(ctx context.Context, libraryID string) (int, error) {
+	type row struct{ ID string }
+	var ids []row
+	q := s.repo.DB.Table("media").Select("id").Where("scrape_status = ?", "matched")
+	if libraryID != "" {
+		q = q.Where("library_id = ?", libraryID)
+	}
+	if err := q.Scan(&ids).Error; err != nil {
+		return 0, err
+	}
+	written := 0
+	for _, r := range ids {
+		if _, err := s.ExportOne(ctx, r.ID); err == nil {
+			written++
+		}
+	}
+	return written, nil
+}
+
+func nfoPath(media string) string {
+	dir := filepath.Dir(media)
+	base := strings.TrimSuffix(filepath.Base(media), filepath.Ext(media))
+	return filepath.Join(dir, fmt.Sprintf("%s.nfo", base))
+}
@@ -4,12 +4,11 @@
 // from one or more providers. Selection is driven by the library type:
 //
 //   library.type == "anime"   -> Bangumi  (fallback: TMDb)
-//   library.type == "tv"      -> TMDb (movies) — TV episodes inherit
-//                                series metadata; episode-level scraping
-//                                is left as a future step
+//   library.type == "tv"      -> TheTVDB  (fallback: TMDb)
 //   default                   -> TMDb
 //
-// The orchestrator publishes scrape progress events on the WS hub.
+// After the primary match we optionally upgrade poster / backdrop with
+// Fanart.tv when an API key is configured.
 package service

 import (
@@ -34,6 +33,8 @@ type ScraperService struct {
 	repo    *repository.Container
 	tmdb    *TMDbProvider
 	bangumi *BangumiProvider
+	thetvdb *TheTVDBProvider
+	fanart  *FanartProvider
 	hub     *Hub
 }

@@ -44,11 +45,13 @@ func NewScraperService(
 	repo *repository.Container,
 	tmdb *TMDbProvider,
 	bangumi *BangumiProvider,
+	thetvdb *TheTVDBProvider,
+	fanart *FanartProvider,
 	hub *Hub,
 ) *ScraperService {
 	return &ScraperService{
 		cfg: cfg, log: log, repo: repo,
-		tmdb: tmdb, bangumi: bangumi, hub: hub,
+		tmdb: tmdb, bangumi: bangumi, thetvdb: thetvdb, fanart: fanart, hub: hub,
 	}
 }

@@ -74,7 +77,6 @@ func CleanQuery(raw string) (title string, year int) {
 	name := strings.TrimSuffix(filepath.Base(raw), filepath.Ext(raw))
 	lower := strings.ToLower(name)

-	// 1. Year first — bracketed years (1999) must survive the next step.
 	if m := yearPattern.FindStringSubmatch(lower); len(m) >= 2 {
 		if v, err := strconv.Atoi(m[1]); err == nil {
 			year = v
@@ -82,10 +84,8 @@ func CleanQuery(raw string) (title string, year int) {
 		}
 	}

-	// 2. Drop everything inside brackets.
 	lower = bracketedTag.ReplaceAllString(lower, " ")

-	// 3. Drop episode markers (S01E02 / 1x02 / EP05 / 第03集).
 	lower = patSEnE.ReplaceAllString(lower, " ")
 	lower = patNxE.ReplaceAllString(lower, " ")
 	lower = patEP.ReplaceAllString(lower, " ")
@@ -102,9 +102,7 @@ func CleanQuery(raw string) (title string, year int) {
 	return strings.TrimSpace(title), year
 }

-// EnrichOne runs the provider chain for a single media row. The library's
-// type decides which provider goes first; a fallback runs when the primary
-// returns nothing.
+// EnrichOne runs the provider chain for a single media row.
 func (s *ScraperService) EnrichOne(ctx context.Context, m *model.Media) error {
 	lib, err := s.repo.Library.FindByID(ctx, m.LibraryID)
 	if err != nil {
@@ -124,12 +122,23 @@ func (s *ScraperService) EnrichOne(ctx context.Context, m *model.Media) error {

 	match := s.lookup(ctx, lib, query, year)
 	if match == nil {
-		// Mark explicitly so we don't retry forever.
 		_ = s.repo.DB.Model(&model.Media{}).Where("id = ?", m.ID).
 			Update("scrape_status", "no_match").Error
 		return nil
 	}

+	// Optional Fanart upgrade.
+	if s.fanart != nil && s.fanart.Enabled() && match.TMDbID > 0 {
+		if a, err := s.fanart.MovieArtwork(ctx, match.TMDbID); err == nil && a != nil {
+			if a.Poster != "" {
+				match.PosterURL = a.Poster
+			}
+			if a.Backdrop != "" {
+				match.BackdropURL = a.Backdrop
+			}
+		}
+	}
+
 	updates := map[string]any{
 		"title":         match.Title,
 		"overview":      match.Overview,
@@ -165,11 +174,19 @@ func (s *ScraperService) lookup(ctx context.Context, lib *model.Library, query s
 	if lib != nil {
 		kind = lib.Type
 	}
-	if kind == "anime" && s.bangumi != nil {
-		if m, err := s.bangumi.Search(ctx, query); err == nil && m != nil {
-			return m
+	switch kind {
+	case "anime":
+		if s.bangumi != nil {
+			if m, err := s.bangumi.Search(ctx, query); err == nil && m != nil {
+				return m
+			}
+		}
+	case "tv":
+		if s.thetvdb != nil && s.thetvdb.Enabled() {
+			if m, err := s.thetvdb.SearchSeries(ctx, query); err == nil && m != nil {
+				return m
+			}
 		}
-		s.log.Debug("bangumi miss, falling back to tmdb", zap.String("query", query))
 	}
 	if s.tmdb != nil && s.tmdb.Enabled() {
 		if m, err := s.tmdb.SearchMovie(ctx, query, year); err == nil && m != nil {
@@ -220,5 +237,8 @@ func (s *ScraperService) AnyEnabled() bool {
 	if s.bangumi != nil && s.bangumi.Enabled() {
 		return true
 	}
+	if s.thetvdb != nil && s.thetvdb.Enabled() {
+		return true
+	}
 	return false
 }
@@ -28,7 +28,10 @@ type Container struct {
 	FFprobe      *FFprobeService
 	TMDb         *TMDbProvider
 	Bangumi      *BangumiProvider
+	TheTVDB      *TheTVDBProvider
+	Fanart       *FanartProvider
 	Scraper      *ScraperService
+	Discover     *DiscoverService
 	Playback     *PlaybackService
 	ImageProxy   *ImageProxy
 	Watcher      *WatcherService
@@ -38,6 +41,8 @@ type Container struct {
 	Stats        *StatsService
 	Profile      *ProfileService
 	Audit        *AuditService
+	NFO          *NFOService
+	AI           *AIService

 	stopCtx    context.Context
 	stopCancel context.CancelFunc
@@ -51,12 +56,17 @@ func New(cfg *config.Config, log *zap.Logger, repos *repository.Container) *Cont
 	probe := NewFFprobeService(cfg, log)
 	tmdb := NewTMDbProvider(cfg, log)
 	bangumi := NewBangumiProvider(cfg, log)
-	scraper := NewScraperService(cfg, log, repos, tmdb, bangumi, hub)
+	thetvdb := NewTheTVDBProvider(cfg, log)
+	fanart := NewFanartProvider(cfg, log)
+	scraper := NewScraperService(cfg, log, repos, tmdb, bangumi, thetvdb, fanart, hub)
+	discover := NewDiscoverService(log, tmdb)
 	transcoder := NewTranscoderService(cfg, log, repos, hub)
 	scanner := NewScannerService(cfg, log, repos, hub, probe, scraper)
 	downloads := NewDownloadService(log, repos, hub)
 	subscription := NewSubscriptionService(log, repos, downloads, hub)
 	watcher := NewWatcherService(log, repos, scanner)
+	nfo := NewNFOService(log, repos)
+	ai := NewAIService(cfg, log)

 	ctx, cancel := context.WithCancel(context.Background())

@@ -73,7 +83,10 @@ func New(cfg *config.Config, log *zap.Logger, repos *repository.Container) *Cont
 		FFprobe:      probe,
 		TMDb:         tmdb,
 		Bangumi:      bangumi,
+		TheTVDB:      thetvdb,
+		Fanart:       fanart,
 		Scraper:      scraper,
+		Discover:     discover,
 		Playback:     NewPlaybackService(log, repos),
 		ImageProxy:   NewImageProxy(cfg, log),
 		Watcher:      watcher,
@@ -83,6 +96,8 @@ func New(cfg *config.Config, log *zap.Logger, repos *repository.Container) *Cont
 		Stats:        NewStatsService(log, repos),
 		Profile:      NewProfileService(log, repos),
 		Audit:        NewAuditService(log, repos),
+		NFO:          nfo,
+		AI:           ai,
 		stopCtx:      ctx,
 		stopCancel:   cancel,
 	}
@@ -0,0 +1,159 @@
+// Package service — TheTVDB v4 provider.
+//
+// TheTVDBProvider implements two methods used by the scraper for TV /
+// anime libraries:
+//
+//   Login()                    -> exchanges secrets.thetvdb_api_key for
+//                                  a JWT (cached for 24h).
+//   SearchSeries(query)        -> /search?query=...&type=series
+//
+// The provider is enabled iff secrets.thetvdb_api_key is non-empty. When
+// disabled every method returns nil, nil so the scraper orchestrator can
+// gracefully fall through.
+package service
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"sync"
+	"time"
+
+	"go.uber.org/zap"
+
+	"github.com/ShukeBta/MediaStationGo/internal/config"
+)
+
+// TheTVDBProvider talks to https://api4.thetvdb.com/v4.
+type TheTVDBProvider struct {
+	cfg    *config.Config
+	log    *zap.Logger
+	client *http.Client
+
+	mu        sync.Mutex
+	token     string
+	tokenExp  time.Time
+}
+
+// NewTheTVDBProvider is the constructor.
+func NewTheTVDBProvider(cfg *config.Config, log *zap.Logger) *TheTVDBProvider {
+	return &TheTVDBProvider{
+		cfg:    cfg,
+		log:    log,
+		client: &http.Client{Timeout: 15 * time.Second},
+	}
+}
+
+// Enabled reports whether an API key is present.
+func (t *TheTVDBProvider) Enabled() bool { return t.cfg.Secrets.TheTVDBAPIKey != "" }
+
+// Login fetches a fresh JWT, cached for 24h.
+func (t *TheTVDBProvider) Login(ctx context.Context) (string, error) {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	if time.Now().Before(t.tokenExp) && t.token != "" {
+		return t.token, nil
+	}
+	body, _ := json.Marshal(map[string]string{"apikey": t.cfg.Secrets.TheTVDBAPIKey})
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost,
+		"https://api4.thetvdb.com/v4/login", bytes.NewReader(body))
+	if err != nil {
+		return "", err
+	}
+	req.Header.Set("Content-Type", "application/json")
+	resp, err := t.client.Do(req)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode >= 400 {
+		return "", fmt.Errorf("thetvdb login: %d", resp.StatusCode)
+	}
+	var out struct {
+		Data struct {
+			Token string `json:"token"`
+		} `json:"data"`
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&out); err != nil {
+		return "", err
+	}
+	t.token = out.Data.Token
+	t.tokenExp = time.Now().Add(24 * time.Hour)
+	return t.token, nil
+}
+
+// SearchSeries returns the top match for a TV / anime query, or nil.
+func (t *TheTVDBProvider) SearchSeries(ctx context.Context, query string) (*Match, error) {
+	if !t.Enabled() || query == "" {
+		return nil, nil
+	}
+	tok, err := t.Login(ctx)
+	if err != nil {
+		return nil, err
+	}
+	u := fmt.Sprintf("https://api4.thetvdb.com/v4/search?query=%s&type=series",
+		urlEscape(query))
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, u, nil)
+	if err != nil {
+		return nil, err
+	}
+	req.Header.Set("Authorization", "Bearer "+tok)
+	resp, err := t.client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer resp.Body.Close()
+	if resp.StatusCode >= 400 {
+		return nil, fmt.Errorf("thetvdb search: %d", resp.StatusCode)
+	}
+
+	type entry struct {
+		ID         string `json:"id"`
+		Name       string `json:"name"`
+		Overview   string `json:"overview"`
+		Image      string `json:"image_url"`
+		Year       string `json:"year"`
+	}
+	type page struct {
+		Data []entry `json:"data"`
+	}
+	var p page
+	if err := json.NewDecoder(resp.Body).Decode(&p); err != nil {
+		return nil, err
+	}
+	if len(p.Data) == 0 {
+		return nil, nil
+	}
+	r := p.Data[0]
+	m := &Match{
+		Title:     r.Name,
+		Overview:  r.Overview,
+		PosterURL: r.Image,
+	}
+	if len(r.Year) >= 4 {
+		fmt.Sscanf(r.Year[:4], "%d", &m.Year)
+	}
+	return m, nil
+}
+
+// urlEscape is a tiny replacement for net/url.QueryEscape kept inline so
+// the file does not pull a second import for one call.
+func urlEscape(s string) string {
+	out := make([]byte, 0, len(s)*3)
+	for _, r := range []byte(s) {
+		switch {
+		case r >= '0' && r <= '9',
+			r >= 'A' && r <= 'Z',
+			r >= 'a' && r <= 'z',
+			r == '-', r == '_', r == '.', r == '~':
+			out = append(out, r)
+		default:
+			out = append(out, '%')
+			const hex = "0123456789ABCDEF"
+			out = append(out, hex[r>>4], hex[r&15])
+		}
+	}
+	return string(out)
+}
@@ -4,16 +4,22 @@
 // into HLS (.m3u8 + .ts). The output lives under cache.cache_dir/hls/<id>.
 // The HTTP layer serves these files directly with a normal http.FileServer.
 //
+// Encoder selection (read once at startup from the config):
+//
+//   transcoder.encoder = "" | "nvenc" | "qsv" | "vaapi"
+//
+//   ""      software libx264 (default; runs anywhere)
+//   nvenc   h264_nvenc      (NVIDIA GPU, requires --gpus all on Docker)
+//   qsv     h264_qsv        (Intel iGPU, requires /dev/dri:/dev/dri)
+//   vaapi   h264_vaapi      (Mesa/Intel VAAPI, requires /dev/dri:/dev/dri
+//                            plus the kernel module loaded)
+//
 // Concurrency model:
 //   - Each Media has at most one active ffmpeg job.
 //   - jobs[mediaID] tracks the running goroutine + cancel func.
 //   - Calling Start while a job already exists is a no-op.
 //   - When the playlist file appears on disk we consider the job "ready"
 //     and unblock the HTTP handler that was waiting on it.
-//
-// The transcode profile is intentionally conservative: a single 720p/1.5M
-// bitrate, AAC stereo audio, MPEG-TS segments. Hardware acceleration
-// (NVENC / QSV / VAAPI) is left as a future config-driven extension point.
 package service

 import (
@@ -34,8 +40,8 @@ import (

 // TranscoderService orchestrates background ffmpeg transcodes.
 type TranscoderService struct {
-	cfg  *config.Config
-	log  *zap.Logger
+	cfg *config.Config
+	log *zap.Logger
 	repo *repository.Container
 	hub  *Hub

@@ -50,6 +56,7 @@ type hlsJob struct {
 	cancel     context.CancelFunc
 	startedAt  time.Time
 	playlistOK bool
+	encoder    string
 }

 // NewTranscoderService is the constructor.
@@ -106,6 +113,7 @@ func (t *TranscoderService) EnsureJob(ctx context.Context, mediaID string) (stri
 		outputDir: outDir,
 		cancel:    cancel,
 		startedAt: time.Now(),
+		encoder:   t.cfg.Transcoder.Encoder,
 	}
 	t.jobs[mediaID] = job
 	t.mu.Unlock()
@@ -158,6 +166,30 @@ func (t *TranscoderService) StopAll() {
 	}
 }

+// ActiveJob is the JSON shape exposed to the React Tasks panel.
+type ActiveJob struct {
+	MediaID    string    `json:"media_id"`
+	Encoder    string    `json:"encoder"`
+	StartedAt  time.Time `json:"started_at"`
+	PlaylistOK bool      `json:"playlist_ok"`
+}
+
+// Active returns a snapshot of the currently running transcode jobs.
+func (t *TranscoderService) Active() []ActiveJob {
+	t.mu.Lock()
+	defer t.mu.Unlock()
+	out := make([]ActiveJob, 0, len(t.jobs))
+	for _, j := range t.jobs {
+		out = append(out, ActiveJob{
+			MediaID:    j.mediaID,
+			Encoder:    j.encoder,
+			StartedAt:  j.startedAt,
+			PlaylistOK: j.playlistOK,
+		})
+	}
+	return out
+}
+
 func (t *TranscoderService) runFFmpeg(ctx context.Context, job *hlsJob, source string) {
 	bin := t.cfg.App.FFmpegPath
 	if bin == "" {
@@ -167,45 +199,19 @@ func (t *TranscoderService) runFFmpeg(ctx context.Context, job *hlsJob, source s
 	playlist := filepath.Join(job.outputDir, "index.m3u8")
 	segments := filepath.Join(job.outputDir, "seg_%05d.ts")

-	args := []string{
-		"-y",
-		"-fflags", "+genpts",
-		"-i", source,
-		// Single 720p H.264 video rendition.
-		"-map", "0:v:0?",
-		"-map", "0:a:0?",
-		"-vf", "scale=-2:min(720\\,ih)",
-		"-c:v", "libx264",
-		"-preset", "veryfast",
-		"-profile:v", "main",
-		"-level", "4.0",
-		"-pix_fmt", "yuv420p",
-		"-b:v", "1500k",
-		"-maxrate", "1800k",
-		"-bufsize", "3000k",
-		"-c:a", "aac",
-		"-ar", "48000",
-		"-b:a", "128k",
-		"-ac", "2",
-		"-force_key_frames", "expr:gte(t,n_forced*4)",
-		"-f", "hls",
-		"-hls_time", "4",
-		"-hls_list_size", "0",
-		"-hls_segment_type", "mpegts",
-		"-hls_flags", "independent_segments",
-		"-hls_segment_filename", segments,
-		playlist,
-	}
+	args := buildFFmpegArgs(t.cfg, source, playlist, segments)

 	cmd := exec.CommandContext(ctx, bin, args...)
 	cmd.Stderr = os.Stderr

 	t.log.Info("transcode started",
 		zap.String("media_id", job.mediaID),
+		zap.String("encoder", job.encoder),
 		zap.String("source", source),
 	)
 	t.hub.Publish("transcode", map[string]any{
 		"media_id": job.mediaID,
+		"encoder":  job.encoder,
 		"status":   "started",
 	})

@@ -227,8 +233,126 @@ func (t *TranscoderService) runFFmpeg(ctx context.Context, job *hlsJob, source s
 	})
 }

+// buildFFmpegArgs assembles the ffmpeg command line for the configured
+// encoder. The function is package-level so the unit test can pin its
+// behaviour without spawning a real ffmpeg process.
+func buildFFmpegArgs(cfg *config.Config, source, playlist, segments string) []string {
+	enc := cfg.Transcoder.Encoder
+	bitrate := cfg.Transcoder.VideoBitrate
+	if bitrate == "" {
+		bitrate = "1500k"
+	}
+	maxrate := cfg.Transcoder.MaxRate
+	if maxrate == "" {
+		maxrate = "1800k"
+	}
+	bufsize := cfg.Transcoder.BufSize
+	if bufsize == "" {
+		bufsize = "3000k"
+	}
+	preset := cfg.Transcoder.Preset
+	if preset == "" {
+		preset = "veryfast"
+	}
+	height := cfg.Transcoder.MaxHeight
+	if height <= 0 {
+		height = 720
+	}
+	segDur := cfg.Transcoder.SegmentSeconds
+	if segDur <= 0 {
+		segDur = 4
+	}
+
+	// Hardware-accel arguments differ in three places:
+	//   - Optional input flags (-hwaccel + device init)
+	//   - Optional input pixel-format upload filter
+	//   - The actual -c:v encoder name + preset/quality flag
+	var pre, vf, vcodec, vpreset string
+	switch enc {
+	case "nvenc":
+		pre = "-hwaccel cuda -hwaccel_output_format cuda"
+		vf = fmt.Sprintf("scale_cuda=-2:min(%d\\,ih)", height)
+		vcodec = "h264_nvenc"
+		vpreset = "p4"
+	case "qsv":
+		pre = "-hwaccel qsv -hwaccel_output_format qsv"
+		vf = fmt.Sprintf("scale_qsv=-1:min(%d\\,ih)", height)
+		vcodec = "h264_qsv"
+		vpreset = preset
+	case "vaapi":
+		device := cfg.App.VAAPIDevice
+		if device == "" {
+			device = "/dev/dri/renderD128"
+		}
+		pre = fmt.Sprintf("-hwaccel vaapi -vaapi_device %s -hwaccel_output_format vaapi", device)
+		vf = fmt.Sprintf("scale_vaapi=-2:min(%d\\,ih),format=nv12|vaapi,hwupload", height)
+		vcodec = "h264_vaapi"
+		vpreset = ""
+	default:
+		// software
+		pre = ""
+		vf = fmt.Sprintf("scale=-2:min(%d\\,ih)", height)
+		vcodec = "libx264"
+		vpreset = preset
+	}
+
+	args := []string{"-y", "-fflags", "+genpts"}
+	for _, p := range splitNonEmptyArgs(pre) {
+		args = append(args, p)
+	}
+	args = append(args, "-i", source, "-map", "0:v:0?", "-map", "0:a:0?", "-vf", vf, "-c:v", vcodec)
+	if vpreset != "" {
+		args = append(args, "-preset", vpreset)
+	}
+	args = append(args,
+		"-pix_fmt", "yuv420p",
+		"-b:v", bitrate,
+		"-maxrate", maxrate,
+		"-bufsize", bufsize,
+		"-c:a", "aac",
+		"-ar", "48000",
+		"-b:a", "128k",
+		"-ac", "2",
+		"-force_key_frames", fmt.Sprintf("expr:gte(t,n_forced*%d)", segDur),
+		"-f", "hls",
+		"-hls_time", fmt.Sprintf("%d", segDur),
+		"-hls_list_size", "0",
+		"-hls_segment_type", "mpegts",
+		"-hls_flags", "independent_segments",
+		"-hls_segment_filename", segments,
+		playlist,
+	)
+	return args
+}
+
+// splitNonEmptyArgs is a tiny helper that mirrors strings.Fields for the
+// pre-input flag string without dragging the strings import into the hot
+// path of every call to buildFFmpegArgs.
+func splitNonEmptyArgs(s string) []string {
+	if s == "" {
+		return nil
+	}
+	out := make([]string, 0, 4)
+	field := make([]rune, 0, 16)
+	flush := func() {
+		if len(field) > 0 {
+			out = append(out, string(field))
+			field = field[:0]
+		}
+	}
+	for _, r := range s {
+		if r == ' ' || r == '\t' {
+			flush()
+			continue
+		}
+		field = append(field, r)
+	}
+	flush()
+	return out
+}
+
 // HumanFFmpegProfile is exposed for the admin UI / settings view.
 func (t *TranscoderService) HumanFFmpegProfile() string {
-	return fmt.Sprintf("ffmpeg=%s, output=%s",
-		t.cfg.App.FFmpegPath, filepath.Join(t.cfg.Cache.CacheDir, "hls"))
+	return fmt.Sprintf("ffmpeg=%s, encoder=%s, output=%s",
+		t.cfg.App.FFmpegPath, t.cfg.Transcoder.Encoder, filepath.Join(t.cfg.Cache.CacheDir, "hls"))
 }
@@ -0,0 +1,46 @@
+package service
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/ShukeBta/MediaStationGo/internal/config"
+)
+
+func TestBuildFFmpegArgs(t *testing.T) {
+	base := &config.Config{}
+	base.Transcoder.MaxHeight = 720
+	base.Transcoder.SegmentSeconds = 4
+	base.App.VAAPIDevice = "/dev/dri/renderD128"
+
+	cases := []struct {
+		name           string
+		encoder        string
+		expectVCodec   string
+		expectInArgs   []string
+		expectNotPresetIfBlank bool
+	}{
+		{"software", "", "libx264", []string{"-preset", "veryfast", "-c:v", "libx264"}, false},
+		{"nvenc", "nvenc", "h264_nvenc", []string{"-hwaccel", "cuda", "-c:v", "h264_nvenc", "-preset", "p4"}, false},
+		{"qsv", "qsv", "h264_qsv", []string{"-hwaccel", "qsv", "-c:v", "h264_qsv"}, false},
+		{"vaapi", "vaapi", "h264_vaapi", []string{"-hwaccel", "vaapi", "-vaapi_device", "/dev/dri/renderD128", "-c:v", "h264_vaapi"}, true},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			cfg := *base
+			cfg.Transcoder.Encoder = tc.encoder
+			args := buildFFmpegArgs(&cfg, "/x.mkv", "/o/x.m3u8", "/o/seg_%05d.ts")
+			joined := strings.Join(args, " ")
+			for _, frag := range tc.expectInArgs {
+				if !strings.Contains(joined, frag) {
+					t.Errorf("expected %q in args, got: %s", frag, joined)
+				}
+			}
+			// vaapi has no -preset flag.
+			if tc.expectNotPresetIfBlank && strings.Contains(joined, "-preset") {
+				t.Errorf("vaapi should not include -preset, got: %s", joined)
+			}
+		})
+	}
+}
@@ -40,8 +40,14 @@ const ProfilePage = lazy(() =>
  import('./pages/ProfilePage').then((m) => ({ default: m.ProfilePage })),
 )
 const StatsPage = lazy(() => import('./pages/StatsPage').then((m) => ({ default: m.StatsPage })))
+const DiscoverPage = lazy(() =>
+  import('./pages/DiscoverPage').then((m) => ({ default: m.DiscoverPage })),
+)
+const TasksPage = lazy(() => import('./pages/TasksPage').then((m) => ({ default: m.TasksPage })))
+const RecycleBinPage = lazy(() =>
+  import('./pages/RecycleBinPage').then((m) => ({ default: m.RecycleBinPage })),
+)

-// Fallback shown while a chunk is loading.
 const Loading = () => <p className="px-6 py-8 text-slate-500">加载中…</p>

 export default function App() {
@@ -59,6 +65,7 @@ export default function App() {
        >
          <Route index element={<HomePage />} />
          <Route path="library/:id" element={<LibraryPage />} />
+          <Route path="discover" element={<DiscoverPage />} />
          <Route path="search" element={<SearchPage />} />
          <Route path="favourites" element={<FavouritesPage />} />
          <Route path="playlists" element={<PlaylistsPage />} />
@@ -68,6 +75,22 @@ export default function App() {
          <Route path="downloads" element={<DownloadsPage />} />
          <Route path="subscriptions" element={<SubscriptionsPage />} />
          <Route path="profile" element={<ProfilePage />} />
+          <Route
+            path="tasks"
+            element={
+              <RequireAdmin>
+                <TasksPage />
+              </RequireAdmin>
+            }
+          />
+          <Route
+            path="recycle"
+            element={
+              <RequireAdmin>
+                <RecycleBinPage />
+              </RequireAdmin>
+            }
+          />
          <Route
            path="stats"
            element={
@@ -0,0 +1,25 @@
+import { api } from './client'
+import type { Media } from '../types'
+
+export interface SearchIntent {
+  query: string
+  year?: number
+  genre?: string
+  type?: string
+  sort?: string
+  language?: string
+}
+
+export const aiAPI = {
+  status: () =>
+    api
+      .get<{ enabled: boolean; provider: string; model: string }>('/ai/status')
+      .then((r) => r.data),
+
+  smartSearch: (query: string) =>
+    api
+      .post<{ intent: SearchIntent; items: Media[] }>('/ai/search', { query })
+      .then((r) => r.data),
+
+  recommend: () => api.get<{ titles: string[] }>('/ai/recommend').then((r) => r.data.titles),
+}
@@ -0,0 +1,21 @@
+import { api } from './client'
+import type { Media } from '../types'
+
+// TMDb-derived "Match" rows used by trending/popular rails. We re-use the
+// Media interface — only TMDb id / poster / overview are populated.
+export interface DiscoverItem extends Partial<Media> {
+  tmdb_id: number
+  title: string
+  overview: string
+  poster_url: string
+  backdrop_url: string
+  year: number
+  rating: number
+}
+
+export const discoverAPI = {
+  trending: () =>
+    api.get<{ items: DiscoverItem[] }>('/discover/trending').then((r) => r.data.items),
+  popular: () =>
+    api.get<{ items: DiscoverItem[] }>('/discover/popular').then((r) => r.data.items),
+}
@@ -0,0 +1,18 @@
+import { api } from './client'
+import type { Media } from '../types'
+
+export const recycleAPI = {
+  list: () => api.get<{ items: Media[] }>('/recycle').then((r) => r.data.items),
+
+  softDelete: (id: string) => api.delete(`/media/${id}`).then((r) => r.data),
+
+  restore: (id: string) => api.post(`/media/${id}/restore`).then((r) => r.data),
+
+  purge: (id: string) => api.delete(`/media/${id}/purge`).then((r) => r.data),
+
+  exportNFO: (id: string) =>
+    api.post<{ path: string }>(`/media/${id}/nfo`).then((r) => r.data),
+
+  exportLibraryNFO: (id: string) =>
+    api.post<{ written: number }>(`/libraries/${id}/nfo`).then((r) => r.data),
+}
@@ -0,0 +1,18 @@
+import { api } from './client'
+import type { QBitTorrent } from '../types'
+
+export interface ActiveTranscode {
+  media_id: string
+  encoder: string
+  started_at: string
+  playlist_ok: boolean
+}
+
+export interface TasksSnapshot {
+  transcodes: ActiveTranscode[]
+  torrents: QBitTorrent[] | null
+}
+
+export const tasksAPI = {
+  snapshot: () => api.get<TasksSnapshot>('/tasks').then((r) => r.data),
+}
@@ -3,14 +3,17 @@ import { Link, NavLink, Outlet, useNavigate } from 'react-router-dom'
 import {
  Activity,
  CloudDownload,
+  Compass,
  Film,
  Heart,
  Home,
+  ListChecks,
  ListMusic,
  LogOut,
  Rss,
  Search,
  Settings,
+  Trash2,
  Library as LibraryIcon,
  User as UserIcon,
 } from 'lucide-react'
@@ -48,6 +51,7 @@ export function Layout() {

        <nav className="flex-1 space-y-1 overflow-y-auto pr-1">
          <SidebarLink to="/" icon={<Home size={18} />} label="首页" end />
+          <SidebarLink to="/discover" icon={<Compass size={18} />} label="发现" />
          <SidebarLink to="/search" icon={<Search size={18} />} label="搜索" />
          <SidebarLink to="/favourites" icon={<Heart size={18} />} label="收藏" />
          <SidebarLink to="/playlists" icon={<ListMusic size={18} />} label="播放列表" />
@@ -83,7 +87,9 @@ export function Layout() {
              <div className="mt-6 px-2 text-xs uppercase tracking-wider text-slate-500">
                管理
              </div>
+              <SidebarLink to="/tasks" icon={<ListChecks size={18} />} label="实时任务" />
              <SidebarLink to="/stats" icon={<Activity size={18} />} label="运行状态" />
+              <SidebarLink to="/recycle" icon={<Trash2 size={18} />} label="回收站" />
              <SidebarLink to="/admin" icon={<Settings size={18} />} label="管理后台" />
            </>
          )}
@@ -0,0 +1,98 @@
+import { useEffect, useState } from 'react'
+import { Sparkles } from 'lucide-react'
+
+import { discoverAPI, type DiscoverItem } from '../api/discover'
+import { imageURL } from '../api/client'
+
+// DiscoverPage shows TMDb trending + popular rails. Rows are empty when
+// no TMDb API key is configured (the backend simply returns []).
+export function DiscoverPage() {
+  const [trending, setTrending] = useState<DiscoverItem[]>([])
+  const [popular, setPopular] = useState<DiscoverItem[]>([])
+  const [loading, setLoading] = useState(true)
+
+  useEffect(() => {
+    Promise.all([
+      discoverAPI.trending().catch(() => [] as DiscoverItem[]),
+      discoverAPI.popular().catch(() => [] as DiscoverItem[]),
+    ])
+      .then(([t, p]) => {
+        setTrending(t)
+        setPopular(p)
+      })
+      .finally(() => setLoading(false))
+  }, [])
+
+  return (
+    <div className="space-y-10">
+      <header className="flex items-center gap-3">
+        <Sparkles className="h-6 w-6 text-primary-400" />
+        <div>
+          <h1 className="font-display text-3xl font-bold text-white">发现</h1>
+          <p className="text-sm text-slate-400">
+            来自 TMDb 的当日热门与流行榜单(需在 secrets.tmdb_api_key 配置 API Key)。
+          </p>
+        </div>
+      </header>
+
+      {loading && <p className="text-slate-500">加载中…</p>}
+
+      {trending.length > 0 && <Row title="今日趋势" items={trending} />}
+      {popular.length > 0 && <Row title="热门电影" items={popular} />}
+      {!loading && trending.length === 0 && popular.length === 0 && (
+        <div className="glass-panel">
+          <p className="text-slate-300">
+            还没有可用的发现数据。前往 <span className="text-primary-400">管理后台 → 设置</span>
+            ,在 secrets.tmdb_api_key 中填写一个有效的 TMDb API Key。
+          </p>
+        </div>
+      )}
+    </div>
+  )
+}
+
+function Row({ title, items }: { title: string; items: DiscoverItem[] }) {
+  return (
+    <section className="space-y-3">
+      <h2 className="font-display text-xl font-semibold text-white">{title}</h2>
+      <div className="grid grid-cols-2 gap-4 sm:grid-cols-3 md:grid-cols-4 lg:grid-cols-5 xl:grid-cols-6">
+        {items.map((m) => (
+          <DiscoverCard key={m.tmdb_id} item={m} />
+        ))}
+      </div>
+    </section>
+  )
+}
+
+function DiscoverCard({ item }: { item: DiscoverItem }) {
+  return (
+    <div className="overflow-hidden rounded-xl border border-white/5 bg-surface-800/60">
+      <div className="aspect-[2/3] w-full bg-surface-900">
+        {item.poster_url ? (
+          <img
+            src={imageURL(item.poster_url)}
+            alt={item.title}
+            loading="lazy"
+            referrerPolicy="no-referrer"
+            className="h-full w-full object-cover"
+          />
+        ) : (
+          <div className="flex h-full w-full items-center justify-center text-slate-600">
+            无海报
+          </div>
+        )}
+      </div>
+      <div className="px-3 py-2">
+        <p className="truncate text-sm font-medium text-white">{item.title}</p>
+        <div className="flex items-center justify-between text-xs text-slate-400">
+          {item.year > 0 && <span>{item.year}</span>}
+          {item.rating > 0 && (
+            <span className="rounded border border-yellow-400/40 px-1.5 py-0.5 text-yellow-400">
+              ★ {item.rating.toFixed(1)}
+            </span>
+          )}
+        </div>
+      </div>
+    </div>
+  )
+}
@@ -1,10 +1,11 @@
 import { useEffect, useState } from 'react'
-import { Link, useParams } from 'react-router-dom'
-import { Heart, Play, RefreshCw, Sparkles } from 'lucide-react'
+import { Link, useNavigate, useParams } from 'react-router-dom'
+import { FileText, Heart, Play, RefreshCw, Sparkles, Trash2 } from 'lucide-react'
 import toast from 'react-hot-toast'

 import { mediaAPI } from '../api/library'
 import { playbackAPI } from '../api/playback'
+import { recycleAPI } from '../api/recycle'
 import { imageURL } from '../api/client'
 import { useAuthStore } from '../stores/auth'
 import { api } from '../api/client'
@@ -39,6 +40,7 @@ function fmtSize(bytes: number): string {
 //   - Re-probe     (admin only, refreshes ffprobe metadata)
 export function MediaDetailPage() {
  const { id = '' } = useParams()
+  const navigate = useNavigate()
  const role = useAuthStore((s) => s.user?.role)
  const [media, setMedia] = useState<Media | null>(null)
  const [favourite, setFavourite] = useState(false)
@@ -81,6 +83,27 @@ export function MediaDetailPage() {
    await refresh()
  }

+  const exportNFO = async () => {
+    if (!media) return
+    try {
+      const r = await recycleAPI.exportNFO(media.id)
+      toast.success(`NFO 已写入 ${r.path}`)
+    } catch (err: unknown) {
+      const msg =
+        (err as { response?: { data?: { error?: string } } })?.response?.data?.error ??
+        '导出失败'
+      toast.error(msg)
+    }
+  }
+
+  const softDelete = async () => {
+    if (!media) return
+    if (!confirm(`将「${media.title}」移至回收站? (磁盘文件保留)`)) return
+    await recycleAPI.softDelete(media.id)
+    toast.success('已移至回收站')
+    navigate(-1)
+  }
+
  if (loading) return <p className="text-slate-500">加载中…</p>
  if (!media) return <p className="text-slate-500">媒体不存在</p>

@@ -149,6 +172,15 @@ export function MediaDetailPage() {
                <button onClick={reprobe} className="neon-button">
                  <RefreshCw size={16} /> 重新探测
                </button>
+                <button onClick={exportNFO} className="neon-button">
+                  <FileText size={16} /> 导出 NFO
+                </button>
+                <button
+                  onClick={softDelete}
+                  className="neon-button !border-red-400/40 !bg-red-400/10 !text-red-400"
+                >
+                  <Trash2 size={16} /> 移至回收站
+                </button>
              </>
            )}
          </div>
@@ -0,0 +1,80 @@
+import { useEffect, useState } from 'react'
+import toast from 'react-hot-toast'
+import { RotateCcw, Trash2 } from 'lucide-react'
+
+import { recycleAPI } from '../api/recycle'
+import type { Media } from '../types'
+
+export function RecycleBinPage() {
+  const [items, setItems] = useState<Media[]>([])
+  const [loading, setLoading] = useState(true)
+
+  const refresh = () =>
+    recycleAPI
+      .list()
+      .then(setItems)
+      .finally(() => setLoading(false))
+
+  useEffect(() => {
+    refresh().catch(() => undefined)
+  }, [])
+
+  return (
+    <div className="space-y-6">
+      <h1 className="font-display text-3xl font-bold text-white">回收站</h1>
+      <p className="text-sm text-slate-400">
+        软删除的媒体保留在数据库中,可以恢复。彻底删除不会移除磁盘上的文件,只会从数据库清除条目。
+      </p>
+
+      {loading && <p className="text-slate-500">加载中…</p>}
+      {!loading && items.length === 0 && <p className="text-slate-400">回收站为空。</p>}
+
+      {items.length > 0 && (
+        <div className="glass-panel">
+          <table className="w-full text-left text-sm">
+            <thead className="text-xs uppercase tracking-wider text-slate-500">
+              <tr>
+                <th className="py-2">标题</th>
+                <th>路径</th>
+                <th className="text-right">操作</th>
+              </tr>
+            </thead>
+            <tbody>
+              {items.map((m) => (
+                <tr key={m.id} className="border-t border-white/5">
+                  <td className="py-2 text-white">{m.title}</td>
+                  <td className="max-w-md truncate text-slate-400" title={m.path}>
+                    {m.path}
+                  </td>
+                  <td className="space-x-2 py-2 text-right">
+                    <button
+                      className="rounded border border-primary-400/40 px-2 py-1 text-xs text-primary-400 hover:bg-primary-400/10"
+                      onClick={async () => {
+                        await recycleAPI.restore(m.id)
+                        toast.success('已恢复')
+                        await refresh()
+                      }}
+                    >
+                      <RotateCcw size={12} />
+                    </button>
+                    <button
+                      className="rounded border border-red-400/40 px-2 py-1 text-xs text-red-400 hover:bg-red-400/10"
+                      onClick={async () => {
+                        if (!confirm(`彻底删除「${m.title}」? (磁盘文件保留)`)) return
+                        await recycleAPI.purge(m.id)
+                        toast.success('已彻底删除')
+                        await refresh()
+                      }}
+                    >
+                      <Trash2 size={12} />
+                    </button>
+                  </td>
+                </tr>
+              ))}
+            </tbody>
+          </table>
+        </div>
+      )}
+    </div>
+  )
+}
@@ -1,36 +1,108 @@
-import { ChangeEvent, useEffect, useState } from 'react'
+import { ChangeEvent, FormEvent, useEffect, useState } from 'react'
+import toast from 'react-hot-toast'
+import { Sparkles } from 'lucide-react'

+import { aiAPI, type SearchIntent } from '../api/ai'
 import { mediaAPI } from '../api/library'
 import { MediaCard } from '../components/MediaCard'
 import type { Media } from '../types'

-// Search-as-you-type with a 300ms debounce.
+// SearchPage runs a fast LIKE search by default, with an optional AI
+// "smart search" toggle that calls /api/ai/search to produce a structured
+// intent before executing the library lookup.
 export function SearchPage() {
  const [q, setQ] = useState('')
  const [items, setItems] = useState<Media[]>([])
  const [loading, setLoading] = useState(false)
+  const [aiOn, setAiOn] = useState(false)
+  const [aiAvailable, setAiAvailable] = useState(false)
+  const [intent, setIntent] = useState<SearchIntent | null>(null)

  useEffect(() => {
+    aiAPI
+      .status()
+      .then((s) => setAiAvailable(s.enabled))
+      .catch(() => setAiAvailable(false))
+  }, [])
+
+  // Fast LIKE search-as-you-type when AI mode is OFF.
+  useEffect(() => {
+    if (aiOn) return
    const t = setTimeout(() => {
      setLoading(true)
      mediaAPI
        .search(q, 60)
-        .then((d) => setItems(d.items))
+        .then((d) => {
+          setItems(d.items)
+          setIntent(null)
+        })
        .finally(() => setLoading(false))
    }, 300)
    return () => clearTimeout(t)
-  }, [q])
+  }, [q, aiOn])
+
+  const onAISubmit = async (e: FormEvent) => {
+    e.preventDefault()
+    if (!q.trim()) return
+    setLoading(true)
+    try {
+      const data = await aiAPI.smartSearch(q)
+      setItems(data.items)
+      setIntent(data.intent)
+    } catch {
+      toast.error('AI 搜索失败')
+    } finally {
+      setLoading(false)
+    }
+  }

  return (
    <div className="space-y-6">
-      <h1 className="font-display text-3xl font-bold text-white">搜索</h1>
-      <input
-        autoFocus
-        className="input-base"
-        placeholder="按标题搜索…"
-        value={q}
-        onChange={(e: ChangeEvent<HTMLInputElement>) => setQ(e.target.value)}
-      />
+      <header className="flex items-center justify-between">
+        <h1 className="font-display text-3xl font-bold text-white">搜索</h1>
+        <button
+          disabled={!aiAvailable}
+          className={
+            'neon-button !px-3 !py-1 !text-xs ' +
+            (aiOn ? '!border-accent-400 !bg-accent-400/20 !text-accent-400' : '')
+          }
+          onClick={() => setAiOn((on) => !on)}
+          title={aiAvailable ? '启用 AI 智能搜索' : '请先在 ai.* 中配置 API Key'}
+        >
+          <Sparkles size={12} /> {aiOn ? 'AI 已开启' : 'AI 智能搜索'}
+        </button>
+      </header>
+
+      {aiOn ? (
+        <form onSubmit={onAISubmit} className="flex gap-2">
+          <input
+            autoFocus
+            className="input-base"
+            placeholder='例如:“2010 年后的科幻电影” / “最近的动漫”'
+            value={q}
+            onChange={(e: ChangeEvent<HTMLInputElement>) => setQ(e.target.value)}
+          />
+          <button type="submit" className="neon-button">
+            搜索
+          </button>
+        </form>
+      ) : (
+        <input
+          autoFocus
+          className="input-base"
+          placeholder="按标题搜索…"
+          value={q}
+          onChange={(e: ChangeEvent<HTMLInputElement>) => setQ(e.target.value)}
+        />
+      )}
+
+      {intent && (
+        <div className="glass-panel !p-3 text-xs text-slate-300">
+          AI 解析:
+          <span className="ml-2 font-mono text-primary-400">{JSON.stringify(intent)}</span>
+        </div>
+      )}
+
      {loading && <p className="text-slate-500">搜索中…</p>}
      <div className="grid grid-cols-2 gap-4 sm:grid-cols-3 md:grid-cols-4 lg:grid-cols-5 xl:grid-cols-6">
        {items.map((m) => (
@@ -0,0 +1,125 @@
+import { useEffect, useState } from 'react'
+import { Activity } from 'lucide-react'
+
+import { tasksAPI, type TasksSnapshot } from '../api/tasks'
+
+function fmtBytes(n: number): string {
+  if (!n || n <= 0) return '0 B'
+  const u = ['B', 'KB', 'MB', 'GB', 'TB']
+  let v = n
+  let i = 0
+  while (v >= 1024 && i < u.length - 1) {
+    v /= 1024
+    i++
+  }
+  return `${v.toFixed(2)} ${u[i]}`
+}
+
+// TasksPage shows everything the backend is doing right now: ffmpeg
+// transcodes + qBittorrent downloads. Refreshes every 3 s.
+export function TasksPage() {
+  const [snap, setSnap] = useState<TasksSnapshot | null>(null)
+
+  useEffect(() => {
+    let cancelled = false
+    const tick = () =>
+      tasksAPI.snapshot().then((s) => {
+        if (!cancelled) setSnap(s)
+      })
+    void tick()
+    const id = window.setInterval(tick, 3_000)
+    return () => {
+      cancelled = true
+      window.clearInterval(id)
+    }
+  }, [])
+
+  if (!snap) return <p className="text-slate-500">加载中…</p>
+
+  const torrents = snap.torrents ?? []
+
+  return (
+    <div className="space-y-8">
+      <header className="flex items-center gap-3">
+        <Activity className="h-6 w-6 text-primary-400" />
+        <h1 className="font-display text-3xl font-bold text-white">实时任务</h1>
+      </header>
+
+      <section className="glass-panel">
+        <h2 className="mb-3 font-display text-lg font-semibold text-white">转码任务</h2>
+        {snap.transcodes.length === 0 && <p className="text-slate-500">暂无运行中转码。</p>}
+        {snap.transcodes.length > 0 && (
+          <table className="w-full text-left text-sm">
+            <thead className="text-xs uppercase tracking-wider text-slate-500">
+              <tr>
+                <th className="py-2">媒体 ID</th>
+                <th>编码器</th>
+                <th>开始时间</th>
+                <th>就绪</th>
+              </tr>
+            </thead>
+            <tbody>
+              {snap.transcodes.map((t) => (
+                <tr key={t.media_id} className="border-t border-white/5">
+                  <td className="py-2 font-mono text-xs text-white">{t.media_id}</td>
+                  <td className="text-slate-300">{t.encoder || 'libx264'}</td>
+                  <td className="text-slate-300">{new Date(t.started_at).toLocaleTimeString()}</td>
+                  <td>
+                    {t.playlist_ok ? (
+                      <span className="rounded border border-emerald-400/40 px-1.5 py-0.5 text-xs text-emerald-400">
+                        ready
+                      </span>
+                    ) : (
+                      <span className="rounded border border-yellow-400/40 px-1.5 py-0.5 text-xs text-yellow-400">
+                        starting
+                      </span>
+                    )}
+                  </td>
+                </tr>
+              ))}
+            </tbody>
+          </table>
+        )}
+      </section>
+
+      <section className="glass-panel">
+        <h2 className="mb-3 font-display text-lg font-semibold text-white">下载任务</h2>
+        {torrents.length === 0 && <p className="text-slate-500">暂无运行中下载。</p>}
+        {torrents.length > 0 && (
+          <table className="w-full text-left text-sm">
+            <thead className="text-xs uppercase tracking-wider text-slate-500">
+              <tr>
+                <th className="py-2">名称</th>
+                <th>状态</th>
+                <th>进度</th>
+                <th>体积</th>
+              </tr>
+            </thead>
+            <tbody>
+              {torrents.map((t) => (
+                <tr key={t.hash} className="border-t border-white/5 align-top">
+                  <td className="max-w-md truncate py-2 text-white" title={t.name}>
+                    {t.name}
+                  </td>
+                  <td className="text-slate-300">{t.state}</td>
+                  <td className="text-slate-300">
+                    <div className="flex items-center gap-2">
+                      <div className="h-1 w-24 overflow-hidden rounded bg-white/10">
+                        <div
+                          className="h-full bg-primary-400"
+                          style={{ width: `${Math.round(t.progress * 100)}%` }}
+                        />
+                      </div>
+                      {(t.progress * 100).toFixed(1)}%
+                    </div>
+                  </td>
+                  <td className="text-slate-300">{fmtBytes(t.size)}</td>
+                </tr>
+              ))}
+            </tbody>
+          </table>
+        )}
+      </section>
+    </div>
+  )
+}