feat(media): images and recordings in backups, exports and imports (M6)

- Backup ZIP (download and automatic per-user server backups) carries the
  files of the user's own notes as media/<id><ext>; restoring creates
  them with new IDs and rewrites the blocks.
- Markdown archive puts them under attachments/ and embeds them relative
  to each note (![caption](../attachments/…)); re-importing turns them
  back into image/audio blocks.
- Obsidian import takes images and audio a note embeds on its own
  (![[file.png]], ![](path/file.png)); unused or invalid files appear as
  skipped in the preview instead of vanishing.
- Format detection and file paths moved into storage for this.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

internal/api/handlers_account.go

@@ -156,7 +156,7 @@ func (s *Server) handleExport(w http.ResponseWriter, r *http.Request) {
 	if r.URL.Query().Get("format") == "zip" {
 		w.Header().Set("Content-Type", "application/zip")
 		w.Header().Set("Content-Disposition", fmt.Sprintf(`attachment; filename="mindsift-backup-%s.zip"`, time.Now().Format("2006-01-02")))
-		if err := storage.WriteBackupZip(exportData, w); err != nil {
+		if err := s.store.WriteBackupZip(userID, exportData, w); err != nil {
 			slog.Error("Backup-Export fehlgeschlagen", "user_id", userID, "error", err)
 		}
 		return

internal/api/media.go

@@ -23,35 +23,9 @@ import (
 
 const mediaGrace = 24 * time.Hour // uploads not yet placed in a block are kept this long
 
-// Accepted formats (detected from the content, not from the name): file extension and kind
-var mediaTypes = map[string]struct{ ext, kind string }{
-	"image/jpeg":      {".jpg", "image"},
-	"image/png":       {".png", "image"},
-	"image/gif":       {".gif", "image"},
-	"image/webp":      {".webp", "image"},
-	"audio/webm":      {".webm", "audio"},
-	"video/webm":      {".webm", "audio"}, // Chrome's audio-only recordings are detected as video/webm
-	"audio/ogg":       {".ogg", "audio"},
-	"application/ogg": {".ogg", "audio"},
-	"audio/mpeg":      {".mp3", "audio"},
-	"audio/wave":      {".wav", "audio"},
-	"audio/wav":       {".wav", "audio"},
-	"audio/mp4":       {".m4a", "audio"},
-	"video/mp4":       {".m4a", "audio"}, // Safari records MP4/AAC
-	"audio/aac":       {".aac", "audio"},
-}
+func (s *Server) mediaRoot() string { return s.store.MediaRoot() }
 
-func (s *Server) mediaRoot() string {
-	return filepath.Join(filepath.Dir(s.cfg.DatabasePath), "media")
-}
-
-func (s *Server) mediaPath(m *storage.Media) string {
-	ext := ".bin"
-	if t, ok := mediaTypes[m.Mime]; ok {
-		ext = t.ext
-	}
-	return filepath.Join(s.mediaRoot(), filepath.Base(m.DocID), filepath.Base(m.ID)+ext)
-}
+func (s *Server) mediaPath(m *storage.Media) string { return s.store.MediaFilePath(m) }
 
 // mediaLimit is the largest allowed file of a kind in bytes.
 func (s *Server) mediaLimit(kind string) int64 {
@@ -65,37 +39,7 @@ func (s *Server) mediaLimit(kind string) int64 {
 	return int64(mb) << 20
 }
 
-// detectMedia recognises the format from the first bytes. Browsers send audio as e.g.
-// "audio/webm;codecs=opus"; that name is kept when the content looks like a container we accept.
-func detectMedia(data []byte, declared string) (mime, kind string, ok bool) {
-	sniffed := http.DetectContentType(data)
-	if i := strings.IndexByte(sniffed, ';'); i >= 0 {
-		sniffed = sniffed[:i]
-	}
-	// Go does not know WebP and M4A by name; check their signatures directly
-	switch {
-	case len(data) >= 12 && string(data[0:4]) == "RIFF" && string(data[8:12]) == "WEBP":
-		sniffed = "image/webp"
-	case len(data) >= 12 && string(data[4:8]) == "ftyp":
-		sniffed = "video/mp4"
-	}
-	t, ok := mediaTypes[sniffed]
-	if !ok {
-		return "", "", false
-	}
-	if t.kind == "audio" {
-		if d := strings.ToLower(strings.TrimSpace(strings.SplitN(declared, ";", 2)[0])); strings.HasPrefix(d, "audio/") {
-			if dt, known := mediaTypes[d]; known && dt.kind == "audio" {
-				sniffed = d
-			}
-		} else if sniffed == "video/webm" {
-			sniffed = "audio/webm"
-		} else if sniffed == "video/mp4" {
-			sniffed = "audio/mp4"
-		}
-	}
-	return sniffed, t.kind, true
-}
+var detectMedia = storage.DetectMedia
 
 func (s *Server) canEditDoc(userID, docID string) bool {
 	role := s.store.DocRole(userID, docID)

internal/storage/backup.go

@@ -123,7 +123,7 @@ func (s *Store) WriteUserBackup(userID, dir string, now time.Time) (string, erro
 		return "", err
 	}
 	name := "mindsift-backup-" + now.Format(backupStampFmt) + ".zip"
-	return name, writeFileAtomically(filepath.Join(dir, name), func(w io.Writer) error { return WriteBackupZip(exp, w) })
+	return name, writeFileAtomically(filepath.Join(dir, name), func(w io.Writer) error { return s.WriteBackupZip(userID, exp, w) })
 }
 
 func gzipFile(src, dst string) error {

internal/storage/export_markdown.go

@@ -32,15 +32,36 @@ func (s *Store) WriteMarkdownVault(userID string, w io.Writer) error {
 		}
 	}
 
+	// Images and recordings go to attachments/, notes link them relative to their folder
+	attachments := map[string]string{} // media ID -> path in the archive
+	if media, err := s.MediaOfUser(userID); err == nil {
+		for _, m := range media {
+			attachments[m.ID] = vaultAttachmentDir + "/" + m.ID + MediaExt(m.Mime)
+		}
+	}
+	if err := s.writeMediaFiles(zw, userID, func(m Media) string { return attachments[m.ID] }); err != nil {
+		return err
+	}
+
 	docs, err := s.GetDocuments(userID)
 	if err != nil {
 		return err
 	}
 	for _, d := range docs {
-		name := names.file(names.dir(d.FolderPath), d.Title, "Unbenannte Notiz")
+		dir := names.dir(d.FolderPath)
+		name := names.file(dir, d.Title, "Unbenannte Notiz")
+		up := ""
+		if dir != "" {
+			up = strings.Repeat("../", strings.Count(dir, "/")+1)
+		}
 		body := markdownFrontmatter(map[string]any{
 			"title": d.Title, "icon": d.Icon, "tags": d.Tags, "created_at": d.CreatedAt, "updated_at": d.UpdatedAt,
-		}) + BlocksToVaultMarkdown(d.Blocks)
+		}) + blocksToVaultMarkdown(d.Blocks, func(id string) string {
+			if p, ok := attachments[id]; ok {
+				return up + p
+			}
+			return ""
+		})
 		if err := writeZipFile(zw, name, body, parseISO(d.UpdatedAt, now)); err != nil {
 			return err
 		}
@@ -164,7 +185,16 @@ func taskMarkdownLine(t domain.Thought) string {
 // BlocksToVaultMarkdown turns outliner blocks into Markdown that Obsidian renders natively:
 // top-level headings, dividers and callouts as real block elements, everything else as a nested
 // list (one tab per level), code as fenced blocks inside the list.
+// vaultAttachmentDir holds images and recordings in the Markdown archive.
+const vaultAttachmentDir = "attachments"
+
 func BlocksToVaultMarkdown(blocks []domain.DocumentBlock) string {
+	return blocksToVaultMarkdown(blocks, nil)
+}
+
+// blocksToVaultMarkdown renders blocks as Markdown; attachment maps a media ID to its relative path
+// (image and audio blocks become embeds Obsidian shows inline).
+func blocksToVaultMarkdown(blocks []domain.DocumentBlock, attachment func(id string) string) string {
 	var sb strings.Builder
 	// Children of a top-level heading/callout/code/divider become a list one level up; otherwise their
 	// tab indent would turn them into an indented code block in Markdown
@@ -178,6 +208,13 @@ func BlocksToVaultMarkdown(blocks []domain.DocumentBlock) string {
 		}
 		indent := strings.Repeat("\t", lvl)
 		content := strings.ReplaceAll(b.Content, "\r\n", "\n")
+		if (b.Type == "image" || b.Type == "audio") && attachment != nil {
+			if ref, ok := ParseMediaContent(content); ok {
+				if p := attachment(ref.ID); p != "" {
+					content = "![" + ref.Caption + "](" + strings.ReplaceAll(p, " ", "%20") + ")"
+				}
+			}
+		}
 		// Soft line breaks continue the list item, aligned under its text
 		cont := func(text, prefix string) string {
 			return strings.ReplaceAll(text, "\n", "\n"+indent+prefix)

internal/storage/import_markdown.go

@@ -7,7 +7,10 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"net/url"
+	"os"
 	"path"
+	"path/filepath"
 	"regexp"
 	"sort"
 	"strings"
@@ -23,6 +26,7 @@ const (
 	maxImportFileBytes     = 10 << 20  // one unpacked note
 	maxImportUnpackedBytes = 500 << 20 // all unpacked notes together
 	maxImportFiles         = 20000
+	maxImportMediaBytes    = 100 << 20 // one image or recording
 )
 
 // ImportPlan is everything an import would create. It is built completely (and validated) before
@@ -35,13 +39,15 @@ type ImportPlan struct {
 	Inbox     []importTask       `json:"-"`
 	Themes    []domain.UserTheme `json:"-"`
 	Skipped   []ImportSkip       `json:"skipped"`
+	// Images and recordings of the archive: "media:<id>" (own backup) or the path in the archive
+	Media map[string][]byte `json:"-"`
 }
 
 // Only keeps the chosen sections of the plan (see NormalizeSections).
 func (p *ImportPlan) Only(sections map[string]bool) *ImportPlan {
 	out := &ImportPlan{Skipped: p.Skipped}
 	if sections[SectionNotes] {
-		out.Folders, out.Docs = p.Folders, p.Docs
+		out.Folders, out.Docs, out.Media = p.Folders, p.Docs, p.Media
 	}
 	if sections[SectionTasks] {
 		out.TaskLists = p.TaskLists
@@ -79,6 +85,7 @@ type ImportSummary struct {
 
 type importDoc struct {
 	folder, title, icon string
+	dir                 string // folder of the note inside the archive, for relative image links
 	blocks              []domain.DocumentBlock
 }
 
@@ -161,7 +168,23 @@ func readZipNotes(r io.ReaderAt, size int64, plan *ImportPlan) ([]zipNote, []str
 			continue
 		}
 		if !strings.EqualFold(path.Ext(name), ".md") {
-			plan.skip(name, "kein Markdown (Anhänge wie Bilder oder PDFs werden nicht importiert)")
+			if !importMediaExt[strings.ToLower(path.Ext(name))] {
+				plan.skip(name, "kein Markdown und kein Bild/Audio (z. B. PDFs werden nicht importiert)")
+				continue
+			}
+			if f.UncompressedSize64 > maxImportMediaBytes {
+				plan.skip(name, "zu groß (max. 100 MB pro Bild oder Aufnahme)")
+				continue
+			}
+			data, err := readZipFile(f, maxImportMediaBytes)
+			if err != nil {
+				plan.skip(name, "nicht lesbar")
+				continue
+			}
+			if unpacked += int64(len(data)); unpacked > maxImportUnpackedBytes {
+				return nil, nil, "", errors.New("Archiv ist entpackt zu groß (max. 500 MB)")
+			}
+			plan.addMedia(name, data)
 			continue
 		}
 		if f.UncompressedSize64 > maxImportFileBytes {
@@ -272,8 +295,9 @@ func PlanObsidianImport(r io.ReaderAt, size int64, fileName string) (*ImportPlan
 		if extra := missingTags(fm.tags, body); extra != "" {
 			blocks = append(blocks, newBlock("bullet", extra, 0))
 		}
-		plan.Docs = append(plan.Docs, importDoc{folder: folder, title: title, icon: fm.icon, blocks: blocks})
+		plan.Docs = append(plan.Docs, importDoc{folder: folder, title: title, icon: fm.icon, dir: path.Dir(n.path), blocks: blocks})
 	}
+	plan.reportUnusedMedia()
 	plan.Folders = sortedKeys(folders)
 	if len(plan.Docs) == 0 {
 		return nil, errors.New("Im Archiv wurden keine Markdown-Notizen gefunden")
@@ -315,9 +339,10 @@ func PlanMindSiftArchiveImport(r io.ReaderAt, size int64) (*ImportPlan, error) {
 			if folder != "" {
 				folders[folder] = true
 			}
-			plan.Docs = append(plan.Docs, importDoc{folder: folder, title: title, icon: fm.icon, blocks: MarkdownToBlocks(body)})
+			plan.Docs = append(plan.Docs, importDoc{folder: folder, title: title, icon: fm.icon, dir: path.Dir(n.path), blocks: MarkdownToBlocks(body)})
 		}
 	}
+	plan.reportUnusedMedia()
 	plan.Folders = sortedKeys(folders)
 	if len(plan.Docs)+len(plan.TaskLists)+len(plan.Lists) == 0 {
 		return nil, errors.New("Das Archiv enthält keine MindSift-Daten (Notizen, Aufgaben oder Listen)")
@@ -328,10 +353,12 @@ func PlanMindSiftArchiveImport(r io.ReaderAt, size int64) (*ImportPlan, error) {
 // BackupFileName is the JSON inside the backup ZIP.
 const BackupFileName = "mindsift-backup.json"
 
-// WriteBackupZip streams the JSON backup compressed into a ZIP with a single file.
-func WriteBackupZip(exp *domain.FullExport, w io.Writer) error {
+// WriteBackupZip streams the JSON backup compressed into a ZIP, plus the images and recordings of the
+// user's notes as media/<id><ext> (blocks point to them as media:<id>).
+func (s *Store) WriteBackupZip(userID string, exp *domain.FullExport, w io.Writer) error {
 	zw := zip.NewWriter(w)
-	f, err := zw.CreateHeader(&zip.FileHeader{Name: BackupFileName, Method: zip.Deflate, Modified: parseISO(exp.ExportedAt, time.Now())})
+	modified := parseISO(exp.ExportedAt, time.Now())
+	f, err := zw.CreateHeader(&zip.FileHeader{Name: BackupFileName, Method: zip.Deflate, Modified: modified})
 	if err != nil {
 		return err
 	}
@@ -340,9 +367,36 @@ func WriteBackupZip(exp *domain.FullExport, w io.Writer) error {
 	if err := enc.Encode(exp); err != nil {
 		return err
 	}
+	if err := s.writeMediaFiles(zw, userID, func(m Media) string { return "media/" + m.ID + MediaExt(m.Mime) }); err != nil {
+		return err
+	}
 	return zw.Close()
 }
 
+// writeMediaFiles adds the user's media files to an archive (already compressed formats: stored as is).
+// Files missing on disk are left out.
+func (s *Store) writeMediaFiles(zw *zip.Writer, userID string, name func(Media) string) error {
+	media, err := s.MediaOfUser(userID)
+	if err != nil {
+		return err
+	}
+	for _, m := range media {
+		src, err := os.Open(s.MediaFilePath(&m))
+		if err != nil {
+			continue
+		}
+		dst, err := zw.CreateHeader(&zip.FileHeader{Name: name(m), Method: zip.Store, Modified: parseISO(m.CreatedAt, time.Now())})
+		if err == nil {
+			_, err = io.Copy(dst, src)
+		}
+		src.Close()
+		if err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
 // PlanMindSiftImport picks the format of MindSift's own data: plain JSON backup, the backup ZIP
 // (one JSON inside) or the Markdown archive.
 func PlanMindSiftImport(data []byte, fileName string) (*ImportPlan, error) {
@@ -367,6 +421,22 @@ func PlanMindSiftImport(data []byte, fileName string) (*ImportPlan, error) {
 	if backup == nil {
 		return PlanMindSiftArchiveImport(bytes.NewReader(data), int64(len(data)))
 	}
+	// Images and recordings next to the JSON: media/<id><ext>
+	media := map[string][]byte{}
+	var mediaBytes int64
+	for _, f := range zr.File {
+		name := cleanArchivePath(f.Name)
+		if !strings.HasPrefix(name, "media/") || f.FileInfo().IsDir() || f.UncompressedSize64 > maxImportMediaBytes {
+			continue
+		}
+		if mediaBytes += int64(f.UncompressedSize64); mediaBytes > maxImportUnpackedBytes {
+			return nil, errors.New("Backup ist entpackt zu groß (max. 500 MB)")
+		}
+		if b, err := readZipFile(f, maxImportMediaBytes); err == nil {
+			base := path.Base(name)
+			media["media:"+strings.TrimSuffix(base, path.Ext(base))] = b
+		}
+	}
 	if backup.UncompressedSize64 > maxImportUnpackedBytes {
 		return nil, errors.New("Backup ist entpackt zu groß (max. 500 MB)")
 	}
@@ -379,7 +449,14 @@ func PlanMindSiftImport(data []byte, fileName string) (*ImportPlan, error) {
 	if err != nil || len(raw) > maxImportUnpackedBytes {
 		return nil, errors.New("Backup im Archiv ist nicht lesbar oder zu groß")
 	}
-	return PlanMindSiftJSONImport(raw)
+	plan, err := PlanMindSiftJSONImport(raw)
+	if err != nil {
+		return nil, err
+	}
+	if len(media) > 0 {
+		plan.Media = media
+	}
+	return plan, nil
 }
 
 // PlanMindSiftJSONImport reads the JSON backup ("Backup herunterladen"). Notes shared with the
@@ -461,7 +538,12 @@ func (s *Store) ApplyImport(userID string, plan *ImportPlan) (ImportSummary, err
 		if icon == "" {
 			icon = "📄"
 		}
-		if _, err := s.CreateDocument(userID, "", title, icon, d.folder, d.blocks, nil, false); err != nil {
+		blocks, files := linkImportMedia(d, plan.Media, nil)
+		doc, err := s.CreateDocument(userID, "", title, icon, d.folder, blocks, nil, false)
+		if err != nil {
+			return plan.Summary(false), fmt.Errorf("Notiz %q: %w", title, err)
+		}
+		if err := s.storeImportMedia(userID, doc.ID, files); err != nil {
 			return plan.Summary(false), fmt.Errorf("Notiz %q: %w", title, err)
 		}
 	}
@@ -902,3 +984,147 @@ func sortedKeys(m map[string]bool) []string {
 	sort.Strings(out)
 	return out
 }
+
+// --- Images and recordings ----------------------------------------------------------------------------
+
+var importMediaExt = map[string]bool{".png": true, ".jpg": true, ".jpeg": true, ".gif": true, ".webp": true,
+	".webm": true, ".ogg": true, ".oga": true, ".mp3": true, ".wav": true, ".m4a": true, ".aac": true}
+
+// reportUnusedMedia lists files no note shows (or that are no valid image/recording) as skipped in the
+// preview and drops them from the plan.
+func (p *ImportPlan) reportUnusedMedia() {
+	if len(p.Media) == 0 {
+		return
+	}
+	used := map[string]bool{}
+	for _, d := range p.Docs {
+		linkImportMedia(d, p.Media, used)
+	}
+	for _, k := range sortedKeys(keySet(p.Media)) {
+		if used[k] {
+			continue
+		}
+		if _, _, ok := DetectMedia(p.Media[k], ""); !ok {
+			p.skip(k, "kein gültiges Bild bzw. keine gültige Aufnahme")
+		} else {
+			p.skip(k, "wird in keiner Notiz angezeigt")
+		}
+		delete(p.Media, k)
+	}
+}
+
+func keySet(m map[string][]byte) map[string]bool {
+	out := make(map[string]bool, len(m))
+	for k := range m {
+		out[k] = true
+	}
+	return out
+}
+
+func (p *ImportPlan) addMedia(name string, data []byte) {
+	if p.Media == nil {
+		p.Media = map[string][]byte{}
+	}
+	p.Media[name] = data
+}
+
+func readZipFile(f *zip.File, limit int64) ([]byte, error) {
+	rc, err := f.Open()
+	if err != nil {
+		return nil, err
+	}
+	defer rc.Close()
+	data, err := io.ReadAll(io.LimitReader(rc, limit+1))
+	if err != nil || int64(len(data)) > limit {
+		return nil, errors.New("nicht lesbar oder zu groß")
+	}
+	return data, nil
+}
+
+type importMediaFile struct {
+	id, kind, mime string
+	data           []byte
+}
+
+var (
+	mdEmbedRe   = regexp.MustCompile(`^!\[([^\]]*)\]\(<?([^)>]+?)>?(?:\s+"[^"]*")?\)$`)
+	wikiEmbedRe = regexp.MustCompile(`^!\[\[([^|\]]+)(?:\|([^\]]*))?\]\]$`)
+	sizeOnlyRe  = regexp.MustCompile(`^\d+(x\d+)?$`)
+)
+
+// linkImportMedia turns blocks that only show a file of the archive into image/audio blocks with new
+// file IDs: media:<id> of an own backup, Markdown images relative to the note, Obsidian ![[file]]
+// embeds (found by name anywhere in the archive). Files without such a block are not imported.
+func linkImportMedia(d importDoc, media map[string][]byte, used map[string]bool) ([]domain.DocumentBlock, []importMediaFile) {
+	if len(media) == 0 {
+		return d.blocks, nil
+	}
+	blocks := make([]domain.DocumentBlock, len(d.blocks))
+	copy(blocks, d.blocks)
+	byKey := map[string]importMediaFile{} // one copy per note, even when shown twice
+	var files []importMediaFile
+	for i, b := range blocks {
+		content := strings.TrimSpace(b.Content)
+		key, caption := "", ""
+		if ref, ok := ParseMediaContent(content); ok {
+			key, caption = "media:"+ref.ID, ref.Caption
+		} else if m := mdEmbedRe.FindStringSubmatch(content); m != nil {
+			target := m[2]
+			if u, err := url.PathUnescape(target); err == nil {
+				target = u
+			}
+			key, caption = path.Clean(path.Join(d.dir, target)), m[1]
+		} else if m := wikiEmbedRe.FindStringSubmatch(content); m != nil {
+			name := strings.TrimSpace(m[1])
+			for k := range media {
+				if path.Base(k) == name || k == name {
+					key = k
+					break
+				}
+			}
+			if !sizeOnlyRe.MatchString(m[2]) {
+				caption = m[2]
+			}
+		}
+		data, ok := media[key]
+		if key == "" || !ok {
+			continue
+		}
+		f, seen := byKey[key]
+		if !seen {
+			if used != nil {
+				used[key] = true
+			}
+			mime, kind, ok := DetectMedia(data, "")
+			if !ok {
+				continue
+			}
+			f = importMediaFile{id: domain.NewID("med"), kind: kind, mime: mime, data: data}
+			byKey[key] = f
+			files = append(files, f)
+		}
+		blocks[i].Type = f.kind
+		blocks[i].Content = MediaContent(f.kind, f.id, caption)
+	}
+	return blocks, files
+}
+
+func (s *Store) storeImportMedia(userID, docID string, files []importMediaFile) error {
+	if len(files) == 0 || s.mediaRoot == "" {
+		return nil
+	}
+	for _, f := range files {
+		m, err := s.createMediaWithID(f.id, docID, userID, f.kind, f.mime, int64(len(f.data)), "")
+		if err != nil {
+			return err
+		}
+		dst := s.MediaFilePath(m)
+		if err := os.MkdirAll(filepath.Dir(dst), 0o755); err != nil {
+			return err
+		}
+		if err := os.WriteFile(dst, f.data, 0o644); err != nil {
+			return err
+		}
+	}
+	return nil
+}

internal/storage/import_markdown_test.go

@@ -204,7 +204,7 @@ func TestImportBackupZip(t *testing.T) {
 	seedAnna(t, store, anna)
 	exp, _ := store.ExportUserData(anna.ID)
 	var buf bytes.Buffer
-	if err := WriteBackupZip(exp, &buf); err != nil {
+	if err := store.WriteBackupZip(anna.ID, exp, &buf); err != nil {
 		t.Fatalf("WriteBackupZip: %v", err)
 	}
 	if buf.Len() == 0 {

internal/storage/media.go

@@ -3,6 +3,10 @@ package storage
 import (
 	"database/sql"
 	"errors"
+	"net/http"
+	"path/filepath"
+	"regexp"
+	"strings"
 	"time"
 
 	"github.com/sandrodach/mindsift/internal/domain"
@@ -26,6 +30,72 @@ type Media struct {
 
 var ErrMediaNotFound = errors.New("Datei nicht gefunden")
 
+// Accepted formats (detected from the content, not from the name): file extension and kind
+var mediaTypes = map[string]struct{ ext, kind string }{
+	"image/jpeg":      {".jpg", "image"},
+	"image/png":       {".png", "image"},
+	"image/gif":       {".gif", "image"},
+	"image/webp":      {".webp", "image"},
+	"audio/webm":      {".webm", "audio"},
+	"video/webm":      {".webm", "audio"}, // Chrome's audio-only recordings are detected as video/webm
+	"audio/ogg":       {".ogg", "audio"},
+	"application/ogg": {".ogg", "audio"},
+	"audio/mpeg":      {".mp3", "audio"},
+	"audio/wave":      {".wav", "audio"},
+	"audio/wav":       {".wav", "audio"},
+	"audio/mp4":       {".m4a", "audio"},
+	"video/mp4":       {".m4a", "audio"}, // Safari records MP4/AAC
+	"audio/aac":       {".aac", "audio"},
+}
+
+// DetectMedia recognises the format from the first bytes. Browsers send audio as e.g.
+// "audio/webm;codecs=opus"; that name is kept when the content looks like a container we accept.
+func DetectMedia(data []byte, declared string) (mime, kind string, ok bool) {
+	sniffed := http.DetectContentType(data)
+	if i := strings.IndexByte(sniffed, ';'); i >= 0 {
+		sniffed = sniffed[:i]
+	}
+	// Go does not know WebP and M4A by name; check their signatures directly
+	switch {
+	case len(data) >= 12 && string(data[0:4]) == "RIFF" && string(data[8:12]) == "WEBP":
+		sniffed = "image/webp"
+	case len(data) >= 12 && string(data[4:8]) == "ftyp":
+		sniffed = "video/mp4"
+	}
+	t, ok := mediaTypes[sniffed]
+	if !ok {
+		return "", "", false
+	}
+	if t.kind == "audio" {
+		if d := strings.ToLower(strings.TrimSpace(strings.SplitN(declared, ";", 2)[0])); strings.HasPrefix(d, "audio/") {
+			if dt, known := mediaTypes[d]; known && dt.kind == "audio" {
+				sniffed = d
+			}
+		} else if sniffed == "video/webm" {
+			sniffed = "audio/webm"
+		} else if sniffed == "video/mp4" {
+			sniffed = "audio/mp4"
+		}
+	}
+	return sniffed, t.kind, true
+}
+
+// MediaExt is the file extension for a stored format.
+func MediaExt(mime string) string {
+	if t, ok := mediaTypes[mime]; ok {
+		return t.ext
+	}
+	return ".bin"
+}
+
+// MediaRoot is the folder with all media files ("" for in-memory databases).
+func (s *Store) MediaRoot() string { return s.mediaRoot }
+
+// MediaFilePath is where the file of m lies.
+func (s *Store) MediaFilePath(m *Media) string {
+	return filepath.Join(s.mediaRoot, filepath.Base(m.DocID), filepath.Base(m.ID)+MediaExt(m.Mime))
+}
+
 func (s *Store) migrateMedia() error {
 	return s.execInTx(
 		stmt(`CREATE TABLE IF NOT EXISTS media (
@@ -44,7 +114,11 @@ func (s *Store) migrateMedia() error {
 }
 
 func (s *Store) CreateMedia(docID, ownerID, kind, mime string, size int64, sourceURL string) (*Media, error) {
-	m := &Media{ID: domain.NewID("med"), DocID: docID, OwnerID: ownerID, Kind: kind, Mime: mime, Size: size,
+	return s.createMediaWithID(domain.NewID("med"), docID, ownerID, kind, mime, size, sourceURL)
+}
+
+func (s *Store) createMediaWithID(id, docID, ownerID, kind, mime string, size int64, sourceURL string) (*Media, error) {
+	m := &Media{ID: id, DocID: docID, OwnerID: ownerID, Kind: kind, Mime: mime, Size: size,
 		SourceURL: sourceURL, CreatedAt: nowUTC()}
 	_, err := s.db.Exec(`INSERT INTO media (id, doc_id, owner_id, kind, mime, size, source_url, created_at) VALUES (?, ?, ?, ?, ?, ?, ?, ?)`,
 		m.ID, m.DocID, m.OwnerID, m.Kind, m.Mime, m.Size, m.SourceURL, m.CreatedAt)
@@ -117,3 +191,47 @@ func (s *Store) UnreferencedMedia(olderThan time.Time) ([]Media, error) {
 	}
 	return out, rows.Err()
 }
+
+// MediaOfUser lists the files shown in the user's own notes (for backups and exports).
+func (s *Store) MediaOfUser(userID string) ([]Media, error) {
+	rows, err := s.db.Query(`SELECT DISTINCT m.id, m.doc_id, m.owner_id, m.kind, m.mime, m.size, m.source_url, m.created_at
+		FROM media m JOIN blocks b ON instr(b.content, 'media:' || m.id) > 0
+		JOIN documents d ON d.id = b.document_id WHERE d.user_id = ?`, userID)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+	var out []Media
+	for rows.Next() {
+		var m Media
+		if rows.Scan(&m.ID, &m.DocID, &m.OwnerID, &m.Kind, &m.Mime, &m.Size, &m.SourceURL, &m.CreatedAt) == nil {
+			out = append(out, m)
+		}
+	}
+	return out, rows.Err()
+}
+
+// MediaRef is a block's pointer to a file: ![caption](media:id) for images, [caption](media:id) for audio.
+type MediaRef struct{ ID, Caption string }
+
+var mediaRefRe = regexp.MustCompile(`^!?\[((?:[^\]\\]|\\.)*)\]\(media:(med_[0-9a-f]+)\)$`)
+var mediaRefEsc = regexp.MustCompile(`\\(.)`)
+
+func ParseMediaContent(content string) (MediaRef, bool) {
+	m := mediaRefRe.FindStringSubmatch(strings.TrimSpace(content))
+	if m == nil {
+		return MediaRef{}, false
+	}
+	return MediaRef{ID: m[2], Caption: mediaRefEsc.ReplaceAllString(m[1], "$1")}, true
+}
+
+// MediaContent builds a block's content for a file (same format as the browser, lib/media.ts).
+func MediaContent(kind, id, caption string) string {
+	caption = strings.Join(strings.Fields(strings.ReplaceAll(caption, "\n", " ")), " ")
+	caption = strings.NewReplacer(`\`, `\\`, `]`, `\]`).Replace(caption)
+	prefix := ""
+	if kind == "image" {
+		prefix = "!"
+	}
+	return prefix + "[" + caption + "](media:" + id + ")"
+}

internal/storage/media_test.go

@@ -0,0 +1,193 @@
+package storage
+
+import (
+	"archive/zip"
+	"bytes"
+	"image"
+	"image/png"
+	"io"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"github.com/sandrodach/mindsift/internal/domain"
+)
+
+func pngBytes(t *testing.T, w int) []byte {
+	t.Helper()
+	var buf bytes.Buffer
+	if err := png.Encode(&buf, image.NewRGBA(image.Rect(0, 0, w, 2))); err != nil {
+		t.Fatal(err)
+	}
+	return buf.Bytes()
+}
+
+// storeTestMedia creates a media row with its file and returns it.
+func storeTestMedia(t *testing.T, store *Store, docID, ownerID string, data []byte) *Media {
+	t.Helper()
+	mime, kind, ok := DetectMedia(data, "")
+	if !ok {
+		t.Fatal("test data not recognised")
+	}
+	m, err := store.CreateMedia(docID, ownerID, kind, mime, int64(len(data)), "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	_ = os.MkdirAll(filepath.Dir(store.MediaFilePath(m)), 0o755)
+	if err := os.WriteFile(store.MediaFilePath(m), data, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	return m
+}
+
+// mediaBlocksOf returns the image/audio blocks of the user's notes and the bytes of their files.
+func mediaBlocksOf(t *testing.T, store *Store, userID string) map[string][]byte {
+	t.Helper()
+	docs, _ := store.GetDocuments(userID)
+	out := map[string][]byte{}
+	for _, d := range docs {
+		for _, b := range d.Blocks {
+			if b.Type != "image" && b.Type != "audio" {
+				continue
+			}
+			ref, ok := ParseMediaContent(b.Content)
+			if !ok {
+				t.Fatalf("block content is no media reference: %q", b.Content)
+			}
+			m, err := store.GetMedia(ref.ID)
+			if err != nil || m.OwnerID != userID || m.DocID != d.ID {
+				t.Fatalf("media row of %s: %+v %v", ref.ID, m, err)
+			}
+			data, err := os.ReadFile(store.MediaFilePath(m))
+			if err != nil {
+				t.Fatal(err)
+			}
+			out[ref.Caption] = data
+		}
+	}
+	return out
+}
+
+func TestBackupZipCarriesMedia(t *testing.T) {
+	store, anna, ben := newImportStore(t)
+	doc, _ := store.CreateDocument(anna.ID, "", "Urlaub", "🏖️", "Reisen", nil, nil, false)
+	img := pngBytes(t, 7)
+	m := storeTestMedia(t, store, doc.ID, anna.ID, img)
+	_, _ = store.CreateDocument(anna.ID, "", "Urlaub", "🏖️", "Reisen", []domain.DocumentBlock{
+		{ID: "blk_t", Type: "bullet", Content: "Strand"},
+		{ID: "blk_i", Type: "image", Content: MediaContent("image", m.ID, "Sonnenuntergang [Süd]")},
+	}, nil, false)
+
+	exp, _ := store.ExportUserData(anna.ID)
+	var buf bytes.Buffer
+	if err := store.WriteBackupZip(anna.ID, exp, &buf); err != nil {
+		t.Fatal(err)
+	}
+	zr, _ := zip.NewReader(bytes.NewReader(buf.Bytes()), int64(buf.Len()))
+	found := false
+	for _, f := range zr.File {
+		found = found || f.Name == "media/"+m.ID+".png"
+	}
+	if !found {
+		t.Fatal("backup ZIP must contain the image")
+	}
+
+	plan, err := PlanMindSiftImport(buf.Bytes(), "backup.zip")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := store.ApplyImport(ben.ID, plan); err != nil {
+		t.Fatal(err)
+	}
+	got := mediaBlocksOf(t, store, ben.ID)
+	if !bytes.Equal(got["Sonnenuntergang [Süd]"], img) {
+		t.Fatalf("restored image missing or different: %v", keysOf(got))
+	}
+}
+
+func TestMarkdownVaultCarriesMedia(t *testing.T) {
+	store, anna, ben := newImportStore(t)
+	doc, _ := store.CreateDocument(anna.ID, "", "Platzhalter", "", "", nil, nil, false)
+	img := pngBytes(t, 5)
+	m := storeTestMedia(t, store, doc.ID, anna.ID, img)
+	_, _ = store.CreateDocument(anna.ID, "", "Plan", "", "Projekte/Haus", []domain.DocumentBlock{
+		{ID: "blk_p", Type: "image", Content: MediaContent("image", m.ID, "Grundriss")},
+	}, nil, false)
+
+	var buf bytes.Buffer
+	if err := store.WriteMarkdownVault(anna.ID, &buf); err != nil {
+		t.Fatal(err)
+	}
+	zr, _ := zip.NewReader(bytes.NewReader(buf.Bytes()), int64(buf.Len()))
+	var note string
+	hasFile := false
+	for _, f := range zr.File {
+		if f.Name == "attachments/"+m.ID+".png" {
+			hasFile = true
+		}
+		if f.Name == "Projekte/Haus/Plan.md" {
+			rc, _ := f.Open()
+			b, _ := io.ReadAll(rc)
+			rc.Close()
+			note = string(b)
+		}
+	}
+	if !hasFile || !strings.Contains(note, "![Grundriss](../../attachments/"+m.ID+".png)") {
+		t.Fatalf("vault: attachment %v, note:\n%s", hasFile, note)
+	}
+
+	plan, err := PlanMindSiftImport(buf.Bytes(), "vault.zip")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := store.ApplyImport(ben.ID, plan); err != nil {
+		t.Fatal(err)
+	}
+	if got := mediaBlocksOf(t, store, ben.ID); !bytes.Equal(got["Grundriss"], img) {
+		t.Fatalf("re-imported image missing: %v", keysOf(got))
+	}
+}
+
+func TestObsidianVaultImagesAreImported(t *testing.T) {
+	store, _, ben := newImportStore(t)
+	img1, img2 := pngBytes(t, 3), pngBytes(t, 4)
+	var buf bytes.Buffer
+	zw := zip.NewWriter(&buf)
+	add := func(name string, data []byte) {
+		w, _ := zw.Create(name)
+		_, _ = w.Write(data)
+	}
+	add("Vault/Reisen/Rom.md", []byte("# Rom\n\n- ![[kolosseum.png|400]]\n- ![Brunnen](bilder/trevi%20brunnen.png)\n- Text mit ![[kolosseum.png]] mittendrin\n"))
+	add("Vault/Anhänge/kolosseum.png", img1)
+	add("Vault/Reisen/bilder/trevi brunnen.png", img2)
+	add("Vault/Anhänge/handbuch.pdf", []byte("%PDF-1.4"))
+	_ = zw.Close()
+
+	plan, err := PlanObsidianImport(bytes.NewReader(buf.Bytes()), int64(buf.Len()), "Vault.zip")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := store.ApplyImport(ben.ID, plan); err != nil {
+		t.Fatal(err)
+	}
+	got := mediaBlocksOf(t, store, ben.ID)
+	if !bytes.Equal(got[""], img1) || !bytes.Equal(got["Brunnen"], img2) || len(got) != 2 {
+		t.Fatalf("obsidian images: %v", keysOf(got))
+	}
+	skippedPDF := false
+	for _, s := range plan.Skipped {
+		skippedPDF = skippedPDF || strings.HasSuffix(s.Path, "handbuch.pdf")
+	}
+	if !skippedPDF {
+		t.Fatal("PDF must be reported as skipped")
+	}
+}
+
+func keysOf(m map[string][]byte) []string {
+	out := make([]string, 0, len(m))
+	for k := range m {
+		out = append(out, k)
+	}
+	return out
+}

internal/storage/sqlite.go

@@ -19,6 +19,8 @@ import (
 type Store struct {
 	db     *sql.DB
 	initMu sync.Mutex
+	// mediaRoot holds the files of images and audio in notes ($DATA_DIR/media), see media.go
+	mediaRoot string
 
 	// OnTaskCreated is told about tasks the store creates by itself (next occurrence of a repeating
 	// task), so the server can push them to connected clients.
@@ -50,6 +52,9 @@ func NewStore(dbPath string) (*Store, error) {
 	db.SetMaxIdleConns(4)
 
 	store := &Store{db: db}
+	if dbPath != ":memory:" && !strings.HasPrefix(dbPath, "file:") {
+		store.mediaRoot = filepath.Join(filepath.Dir(dbPath), "media")
+	}
 	if err := store.migrate(); err != nil {
 		return nil, fmt.Errorf("failed to run migrations: %w", err)
 	}