diff --git a/backend/.gitignore b/backend/.gitignore
index 4b42251..2b2b48f 100644
--- a/backend/.gitignore
+++ b/backend/.gitignore
@@ -23,6 +23,7 @@ eggs/
 .eggs/
 lib/
 lib64/
+!app/lib/
 parts/
 sdist/
 var/
diff --git a/backend/app/api/v1/routes_glossaries.py b/backend/app/api/v1/routes_glossaries.py
new file mode 100644
index 0000000..5eaf993
--- /dev/null
+++ b/backend/app/api/v1/routes_glossaries.py
@@ -0,0 +1,288 @@
+"""
+Glossary management endpoints.
+
+Access:
+  - All glossary mutations (upload, activate, archive) → Admin or PM of the client
+  - Glossary reads (list, detail, terms)               → Admin, PM, or staff members
+
+Routes are nested under /clients/{client_id}/glossaries to keep ownership clear.
+"""
+from __future__ import annotations
+
+from fastapi import APIRouter, Depends, File, Form, HTTPException, Query, UploadFile
+
+from ...core.dependencies import get_current_user, require_pm_for_client, require_roles
+from ...core.logging import get_logger
+from ...models.audit_log import AuditAction
+from ...models.glossary import (
+    GlossaryDetailResponse,
+    GlossaryResponse,
+    GlossaryVersionResponse,
+)
+from ...models.user import User, UserRole
+from ...services import audit_logger as audit_svc
+from ...services import glossary_service as svc
+
+logger = get_logger(__name__)
+
+router = APIRouter(
+    prefix="/clients/{client_id}/glossaries",
+    tags=["glossaries"],
+)
+
+_ALLOWED_CONTENT_TYPES = {
+    "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+    "application/vnd.ms-excel",
+}
+_MAX_FILE_SIZE_MB = 50
+
+
+def _require_client_staff(client_id: str):
+    """Dependency: admin or PM of this client."""
+    return require_pm_for_client(client_id_param="client_id")
+
+
+# ── List glossaries ───────────────────────────────────────────────────────────
+
+@router.get("", response_model=list[GlossaryResponse])
+async def list_glossaries(
+    client_id: str,
+    current_user: User = Depends(get_current_user),
+):
+    """List all active glossaries for a client."""
+    _assert_can_read(current_user)
+    glossaries = await svc.get_glossaries_for_client(client_id)
+    return [_to_response(g) for g in glossaries]
+
+
+# ── Upload new glossary ───────────────────────────────────────────────────────
+
+@router.post("", response_model=GlossaryDetailResponse, status_code=201)
+async def upload_glossary(
+    client_id: str,
+    file: UploadFile = File(..., description="xlsx glossary file"),
+    name: str = Form(...),
+    source_locale: str = Form(..., description="BCP-47 source locale, e.g. en-GB"),
+    source_locale_col: str = Form(..., description="xlsx column header for the source language, e.g. en_gb"),
+    description: str | None = Form(None),
+    change_note: str | None = Form(None),
+    current_user: User = Depends(require_roles(UserRole.ADMIN, UserRole.PROJECT_MANAGER)),
+):
+    """Upload a new glossary xlsx file and associate it with a client."""
+    _validate_xlsx(file)
+
+    try:
+        glossary, version = await svc.ingest_glossary(
+            client_id=client_id,
+            name=name,
+            source_locale=source_locale,
+            source_locale_col=source_locale_col,
+            file=file,
+            user_id=str(current_user.id),
+            description=description,
+            change_note=change_note,
+        )
+    except ValueError as exc:
+        raise HTTPException(status_code=422, detail=str(exc)) from exc
+
+    await audit_svc.audit_logger.log_action(
+        action=AuditAction.GLOSSARY_UPLOAD,
+        description=f"Glossary '{name}' uploaded for client {client_id}",
+        user=current_user,
+        resource_type="glossary",
+        resource_id=glossary.id,
+        details={"term_count": version.term_count, "source_locale": source_locale},
+    )
+
+    versions = await svc.get_versions(glossary.id)
+    return _to_detail_response(glossary, versions)
+
+
+# ── Get glossary detail ───────────────────────────────────────────────────────
+
+@router.get("/{glossary_id}", response_model=GlossaryDetailResponse)
+async def get_glossary(
+    client_id: str,
+    glossary_id: str,
+    current_user: User = Depends(get_current_user),
+):
+    _assert_can_read(current_user)
+    glossary = await svc.get_glossary(glossary_id)
+    if not glossary or glossary.client_id != client_id:
+        raise HTTPException(status_code=404, detail="Glossary not found")
+    versions = await svc.get_versions(glossary_id)
+    return _to_detail_response(glossary, versions)
+
+
+# ── Browse terms ──────────────────────────────────────────────────────────────
+
+@router.get("/{glossary_id}/terms")
+async def list_terms(
+    client_id: str,
+    glossary_id: str,
+    version_id: str | None = Query(None, description="Specific version; defaults to active"),
+    search: str | None = Query(None),
+    page: int = Query(1, ge=1),
+    page_size: int = Query(50, ge=1, le=200),
+    current_user: User = Depends(get_current_user),
+):
+    _assert_can_read(current_user)
+    glossary = await svc.get_glossary(glossary_id)
+    if not glossary or glossary.client_id != client_id:
+        raise HTTPException(status_code=404, detail="Glossary not found")
+
+    vid = version_id or glossary.current_version_id
+    if not vid:
+        return {"terms": [], "total": 0, "page": page, "page_size": page_size}
+
+    terms, total = await svc.get_terms_page(vid, search=search, page=page, page_size=page_size)
+    return {
+        "terms": [{"source_term": t.source_term, "translations": t.translations} for t in terms],
+        "total": total,
+        "page": page,
+        "page_size": page_size,
+    }
+
+
+# ── Upload new version ────────────────────────────────────────────────────────
+
+@router.post("/{glossary_id}/versions", response_model=GlossaryVersionResponse, status_code=201)
+async def upload_version(
+    client_id: str,
+    glossary_id: str,
+    file: UploadFile = File(...),
+    source_locale_col: str = Form(...),
+    change_note: str | None = Form(None),
+    current_user: User = Depends(require_roles(UserRole.ADMIN, UserRole.PROJECT_MANAGER)),
+):
+    """Upload a new xlsx file as a new version of an existing glossary."""
+    _validate_xlsx(file)
+    glossary = await svc.get_glossary(glossary_id)
+    if not glossary or glossary.client_id != client_id:
+        raise HTTPException(status_code=404, detail="Glossary not found")
+
+    try:
+        version = await svc.ingest_new_version(
+            glossary_id=glossary_id,
+            source_locale_col=source_locale_col,
+            file=file,
+            user_id=str(current_user.id),
+            change_note=change_note,
+        )
+    except ValueError as exc:
+        raise HTTPException(status_code=422, detail=str(exc)) from exc
+
+    await audit_svc.audit_logger.log_action(
+        action=AuditAction.GLOSSARY_VERSION_UPLOAD,
+        description=f"New glossary version uploaded for glossary {glossary_id}",
+        user=current_user,
+        resource_type="glossary_version",
+        resource_id=version.id,
+        details={"term_count": version.term_count, "version_number": version.version_number},
+    )
+    return _version_to_response(version)
+
+
+# ── Activate a version ────────────────────────────────────────────────────────
+
+@router.post("/{glossary_id}/activate")
+async def activate_version(
+    client_id: str,
+    glossary_id: str,
+    version_id: str = Form(...),
+    current_user: User = Depends(require_roles(UserRole.ADMIN, UserRole.PROJECT_MANAGER)),
+):
+    glossary = await svc.get_glossary(glossary_id)
+    if not glossary or glossary.client_id != client_id:
+        raise HTTPException(status_code=404, detail="Glossary not found")
+
+    try:
+        await svc.activate_version(glossary_id, version_id)
+    except ValueError as exc:
+        raise HTTPException(status_code=404, detail=str(exc)) from exc
+
+    await audit_svc.audit_logger.log_action(
+        action=AuditAction.GLOSSARY_ACTIVATE,
+        description=f"Glossary version {version_id} activated",
+        user=current_user,
+        resource_type="glossary",
+        resource_id=glossary_id,
+        details={"version_id": version_id},
+    )
+    return {"status": "ok", "active_version_id": version_id}
+
+
+# ── Archive (soft-delete) ─────────────────────────────────────────────────────
+
+@router.delete("/{glossary_id}", status_code=204)
+async def archive_glossary(
+    client_id: str,
+    glossary_id: str,
+    current_user: User = Depends(require_roles(UserRole.ADMIN)),
+):
+    glossary = await svc.get_glossary(glossary_id)
+    if not glossary or glossary.client_id != client_id:
+        raise HTTPException(status_code=404, detail="Glossary not found")
+    await svc.archive_glossary(glossary_id)
+    await audit_svc.audit_logger.log_action(
+        action=AuditAction.GLOSSARY_ARCHIVE,
+        description=f"Glossary {glossary_id} archived",
+        user=current_user,
+        resource_type="glossary",
+        resource_id=glossary_id,
+    )
+
+
+# ── Helpers ───────────────────────────────────────────────────────────────────
+
+def _assert_can_read(user: User) -> None:
+    allowed = {UserRole.ADMIN, UserRole.PROJECT_MANAGER, UserRole.REVIEWER,
+               UserRole.LINGUIST, UserRole.PRODUCTION}
+    if user.role not in allowed:
+        raise HTTPException(status_code=403, detail="Insufficient permissions")
+
+
+def _validate_xlsx(file: UploadFile) -> None:
+    if file.content_type not in _ALLOWED_CONTENT_TYPES and not (
+        file.filename and file.filename.endswith(".xlsx")
+    ):
+        raise HTTPException(
+            status_code=422,
+            detail="Only .xlsx files are accepted",
+        )
+
+
+def _to_response(g) -> GlossaryResponse:
+    return GlossaryResponse(
+        id=str(g.id),
+        client_id=g.client_id,
+        name=g.name,
+        description=g.description,
+        source_locale=g.source_locale,
+        source=g.source,
+        status=g.status,
+        current_version_id=g.current_version_id,
+        created_at=g.created_at,
+        created_by=g.created_by,
+    )
+
+
+def _version_to_response(v) -> GlossaryVersionResponse:
+    return GlossaryVersionResponse(
+        id=str(v.id),
+        glossary_id=v.glossary_id,
+        version_number=v.version_number,
+        term_count=v.term_count,
+        embedded_count=v.embedded_count,
+        embedding_status=v.embedding_status,
+        created_at=v.created_at,
+        created_by=v.created_by,
+        change_note=v.change_note,
+    )
+
+
+def _to_detail_response(glossary, versions) -> GlossaryDetailResponse:
+    return GlossaryDetailResponse(
+        **_to_response(glossary).model_dump(),
+        versions=[_version_to_response(v) for v in versions],
+    )
diff --git a/backend/app/core/config.py b/backend/app/core/config.py
index 4a4576c..788188e 100644
--- a/backend/app/core/config.py
+++ b/backend/app/core/config.py
@@ -93,7 +93,24 @@ class Settings(BaseSettings):
         "sv": "sv-SE",
         "es-419": "es-US",
         "pt-BR": "pt-BR",
-        "fr-CA": "fr-CA"
+        "fr-CA": "fr-CA",
+        # Explicit region variants (added for locale-aware glossary support)
+        "de-DE": "de-DE",
+        "en-US": "en-US",
+        "en-GB": "en-GB",
+        "en-CA": "en-CA",
+        "es-ES": "es-ES",
+        "es-MX": "es-US",
+        "fr-FR": "fr-FR",
+        "it-IT": "it-IT",
+        "ja-JP": "ja-JP",
+        "ko-KR": "ko-KR",
+        "nl-NL": "nl-NL",
+        "pl-PL": "pl-PL",
+        "cs-CZ": "cs-CZ",
+        "tr-TR": "tr-TR",
+        "id-ID": "id-ID",
+        "pt-PT": "pt-PT",
     }
     gemini_tts_language_names: dict[str, str] = {
         "en": "English",
@@ -129,7 +146,24 @@ class Settings(BaseSettings):
         "sv": "Swedish",
         "es-419": "Spanish (Latin America)",
         "pt-BR": "Portuguese (Brazil)",
-        "fr-CA": "French (Canada)"
+        "fr-CA": "French (Canada)",
+        # Explicit region variants
+        "de-DE": "German (Germany)",
+        "en-US": "English (US)",
+        "en-GB": "English (UK)",
+        "en-CA": "English (Canada)",
+        "es-ES": "Spanish (Spain)",
+        "es-MX": "Spanish (Mexico)",
+        "fr-FR": "French (France)",
+        "it-IT": "Italian (Italy)",
+        "ja-JP": "Japanese (Japan)",
+        "ko-KR": "Korean (Korea)",
+        "nl-NL": "Dutch (Netherlands)",
+        "pl-PL": "Polish (Poland)",
+        "cs-CZ": "Czech (Czech Republic)",
+        "tr-TR": "Turkish (Turkey)",
+        "id-ID": "Indonesian (Indonesia)",
+        "pt-PT": "Portuguese (Portugal)",
     }
     gemini_tts_preview_samples: dict[str, str] = {
         "en": "This is a preview of the audio description voice.",
@@ -165,7 +199,24 @@ class Settings(BaseSettings):
         "sv": "Det här är en förhandsgranskning av ljudbeskrivningsrösten.",
         "es-419": "Esta es una vista previa de la voz de audiodescripción.",
         "pt-BR": "Esta é uma prévia da voz da audiodescrição.",
-        "fr-CA": "Ceci est un aperçu de la voix de l'audiodescription."
+        "fr-CA": "Ceci est un aperçu de la voix de l'audiodescription.",
+        # Explicit region variants
+        "de-DE": "Dies ist eine Vorschau der Audiodeskriptionsstimme.",
+        "en-US": "This is a preview of the audio description voice.",
+        "en-GB": "This is a preview of the audio description voice.",
+        "en-CA": "This is a preview of the audio description voice.",
+        "es-ES": "Esta es una vista previa de la voz de audiodescripción.",
+        "es-MX": "Esta es una vista previa de la voz de audiodescripción.",
+        "fr-FR": "Ceci est un aperçu de la voix de l'audiodescription.",
+        "it-IT": "Questa è un'anteprima della voce dell'audiodescrizione.",
+        "ja-JP": "これは音声解説の声のプレビューです。",
+        "ko-KR": "이것은 오디오 설명 음성의 미리보기입니다.",
+        "nl-NL": "Dit is een voorbeeld van de audiodescriptiestem.",
+        "pl-PL": "To jest podgląd głosu audiodeskrypcji.",
+        "cs-CZ": "Toto je náhled hlasu zvukového popisu.",
+        "tr-TR": "Bu, sesli betimleme sesinin bir önizlemesidir.",
+        "id-ID": "Ini adalah pratinjau suara deskripsi audio.",
+        "pt-PT": "Esta é uma pré-visualização da voz da audiodescrição.",
     }
 
     # Gemini TTS Model Options
diff --git a/backend/app/lib/locales.py b/backend/app/lib/locales.py
new file mode 100644
index 0000000..62d4013
--- /dev/null
+++ b/backend/app/lib/locales.py
@@ -0,0 +1,245 @@
+"""
+Central locale registry.
+
+Provides a single source of truth for BCP-47 codes, display names,
+and Gemini-friendly labels used throughout the translation/TTS pipeline.
+
+Convention: BCP-47 with hyphen separator (fr-FR, en-GB, pt-BR).
+xlsx underscore format (fr_fr, en_gb) is normalized at import time.
+Bare language-only codes (fr, en) remain valid for legacy compat.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class Locale:
+    code: str            # canonical BCP-47 (e.g. "fr-FR")
+    display_name: str    # human-readable (e.g. "French (France)")
+    gemini_label: str    # what to pass to Gemini prompts (e.g. "French (France)")
+    tts_lang: str        # BCP-47 for TTS API (may differ, e.g. es-MX → es-US)
+    preview_sample: str  # sample sentence for TTS preview
+
+
+# Master locale registry. Bare language codes (legacy) + explicit region variants.
+_REGISTRY: dict[str, Locale] = {loc.code: loc for loc in [
+    # ── English ──────────────────────────────────────────────────────────────
+    Locale("en", "English", "English", "en-US",
+           "This is a preview of the audio description voice."),
+    Locale("en-US", "English (US)", "English (United States)", "en-US",
+           "This is a preview of the audio description voice."),
+    Locale("en-GB", "English (UK)", "English (United Kingdom)", "en-GB",
+           "This is a preview of the audio description voice."),
+    Locale("en-CA", "English (Canada)", "English (Canada)", "en-CA",
+           "This is a preview of the audio description voice."),
+    # ── Spanish ──────────────────────────────────────────────────────────────
+    Locale("es", "Spanish", "Spanish", "es-US",
+           "Esta es una vista previa de la voz de audiodescripcion."),
+    Locale("es-ES", "Spanish (Spain)", "Spanish (Spain)", "es-ES",
+           "Esta es una vista previa de la voz de audiodescripción."),
+    Locale("es-MX", "Spanish (Mexico)", "Spanish (Mexico)", "es-US",
+           "Esta es una vista previa de la voz de audiodescripción."),
+    Locale("es-419", "Spanish (Latin America)", "Spanish (Latin America)", "es-US",
+           "Esta es una vista previa de la voz de audiodescripción."),
+    # ── French ───────────────────────────────────────────────────────────────
+    Locale("fr", "French", "French", "fr-FR",
+           "Ceci est un apercu de la voix de l'audiodescription."),
+    Locale("fr-FR", "French (France)", "French (France)", "fr-FR",
+           "Ceci est un aperçu de la voix de l'audiodescription."),
+    Locale("fr-CA", "French (Canada)", "French (Canada)", "fr-CA",
+           "Ceci est un aperçu de la voix de l'audiodescription."),
+    # ── German ───────────────────────────────────────────────────────────────
+    Locale("de", "German", "German", "de-DE",
+           "Dies ist eine Vorschau der Audiodeskriptionsstimme."),
+    Locale("de-DE", "German (Germany)", "German (Germany)", "de-DE",
+           "Dies ist eine Vorschau der Audiodeskriptionsstimme."),
+    # ── Italian ──────────────────────────────────────────────────────────────
+    Locale("it", "Italian", "Italian", "it-IT",
+           "Questa e un'anteprima della voce dell'audiodescrizione."),
+    Locale("it-IT", "Italian (Italy)", "Italian (Italy)", "it-IT",
+           "Questa è un'anteprima della voce dell'audiodescrizione."),
+    # ── Portuguese ───────────────────────────────────────────────────────────
+    Locale("pt", "Portuguese", "Portuguese", "pt-BR",
+           "Esta e uma previa da voz da audiodescricao."),
+    Locale("pt-BR", "Portuguese (Brazil)", "Portuguese (Brazil)", "pt-BR",
+           "Esta é uma prévia da voz da audiodescrição."),
+    Locale("pt-PT", "Portuguese (Portugal)", "Portuguese (Portugal)", "pt-PT",
+           "Esta é uma pré-visualização da voz da audiodescrição."),
+    # ── Japanese ─────────────────────────────────────────────────────────────
+    Locale("ja", "Japanese", "Japanese", "ja-JP",
+           "これは音声解説の声のプレビューです。"),
+    Locale("ja-JP", "Japanese (Japan)", "Japanese (Japan)", "ja-JP",
+           "これは音声解説の声のプレビューです。"),
+    # ── Korean ───────────────────────────────────────────────────────────────
+    Locale("ko", "Korean", "Korean", "ko-KR",
+           "이것은 오디오 설명 음성의 미리보기입니다."),
+    Locale("ko-KR", "Korean (Korea)", "Korean (South Korea)", "ko-KR",
+           "이것은 오디오 설명 음성의 미리보기입니다."),
+    # ── Arabic ───────────────────────────────────────────────────────────────
+    Locale("ar", "Arabic", "Arabic", "ar-EG",
+           "هذه معاينة لصوت الوصف الصوتي."),
+    # ── Hindi ────────────────────────────────────────────────────────────────
+    Locale("hi", "Hindi", "Hindi", "hi-IN",
+           "यह ऑडियो विवरण आवाज का पूर्वावलोकन है।"),
+    # ── Indonesian ───────────────────────────────────────────────────────────
+    Locale("id", "Indonesian", "Indonesian", "id-ID",
+           "Ini adalah pratinjau suara deskripsi audio."),
+    Locale("id-ID", "Indonesian (Indonesia)", "Indonesian (Indonesia)", "id-ID",
+           "Ini adalah pratinjau suara deskripsi audio."),
+    # ── Dutch ────────────────────────────────────────────────────────────────
+    Locale("nl", "Dutch", "Dutch", "nl-NL",
+           "Dit is een voorbeeld van de audiodescriptiestem."),
+    Locale("nl-NL", "Dutch (Netherlands)", "Dutch (Netherlands)", "nl-NL",
+           "Dit is een voorbeeld van de audiodescriptiestem."),
+    # ── Polish ───────────────────────────────────────────────────────────────
+    Locale("pl", "Polish", "Polish", "pl-PL",
+           "To jest podglad glosu audiodeskrypcji."),
+    Locale("pl-PL", "Polish (Poland)", "Polish (Poland)", "pl-PL",
+           "To jest podgląd głosu audiodeskrypcji."),
+    # ── Russian ──────────────────────────────────────────────────────────────
+    Locale("ru", "Russian", "Russian", "ru-RU",
+           "Это предварительный просмотр голоса аудиоописания."),
+    # ── Thai ─────────────────────────────────────────────────────────────────
+    Locale("th", "Thai", "Thai", "th-TH",
+           "นี่คือตัวอย่างเสียงบรรยายภาพ"),
+    # ── Turkish ──────────────────────────────────────────────────────────────
+    Locale("tr", "Turkish", "Turkish", "tr-TR",
+           "Bu, sesli betimleme sesinin bir onizlemesidir."),
+    Locale("tr-TR", "Turkish (Turkey)", "Turkish (Turkey)", "tr-TR",
+           "Bu, sesli betimleme sesinin bir önizlemesidir."),
+    # ── Vietnamese ───────────────────────────────────────────────────────────
+    Locale("vi", "Vietnamese", "Vietnamese", "vi-VN",
+           "Day la ban xem truoc giong mo ta am thanh."),
+    # ── Romanian ─────────────────────────────────────────────────────────────
+    Locale("ro", "Romanian", "Romanian", "ro-RO",
+           "Aceasta este o previzualizare a vocii descrierii audio."),
+    # ── Ukrainian ────────────────────────────────────────────────────────────
+    Locale("uk", "Ukrainian", "Ukrainian", "uk-UA",
+           "Це попередній перегляд голосу аудіоопису."),
+    # ── Bengali ──────────────────────────────────────────────────────────────
+    Locale("bn", "Bengali", "Bengali", "bn-BD",
+           "এটি অডিও বর্ণনা ভয়েসের একটি প্রিভিউ।"),
+    # ── Marathi ──────────────────────────────────────────────────────────────
+    Locale("mr", "Marathi", "Marathi", "mr-IN",
+           "हे ऑडिओ वर्णन आवाजाचे पूर्वावलोकन आहे."),
+    # ── Tamil ────────────────────────────────────────────────────────────────
+    Locale("ta", "Tamil", "Tamil", "ta-IN",
+           "இது ஆடியோ விளக்க குரலின் முன்னோட்டம்."),
+    # ── Telugu ───────────────────────────────────────────────────────────────
+    Locale("te", "Telugu", "Telugu", "te-IN",
+           "ఇది ఆడియో వివరణ స్వరం యొక్క ప్రివ్యూ."),
+    # ── Chinese ──────────────────────────────────────────────────────────────
+    Locale("zh", "Chinese", "Chinese (Simplified)", "zh-CN",
+           "这是音频描述语音的预览。"),
+    # ── Czech ────────────────────────────────────────────────────────────────
+    Locale("cs", "Czech", "Czech", "cs-CZ",
+           "Toto je náhled hlasu zvukového popisu."),
+    Locale("cs-CZ", "Czech (Czech Republic)", "Czech (Czech Republic)", "cs-CZ",
+           "Toto je náhled hlasu zvukového popisu."),
+    # ── Danish ───────────────────────────────────────────────────────────────
+    Locale("da", "Danish", "Danish", "da-DK",
+           "Dette er en forhåndsvisning af lydbeskrivelsesstemmen."),
+    # ── Finnish ──────────────────────────────────────────────────────────────
+    Locale("fi", "Finnish", "Finnish", "fi-FI",
+           "Tämä on äänikuvauksen äänen esikatselu."),
+    # ── Hungarian ────────────────────────────────────────────────────────────
+    Locale("hu", "Hungarian", "Hungarian", "hu-HU",
+           "Ez a hangos leírás hangjának előnézete."),
+    # ── Norwegian ────────────────────────────────────────────────────────────
+    Locale("no", "Norwegian", "Norwegian", "nb-NO",
+           "Dette er en forhåndsvisning av lydbeskrivelsesstemmen."),
+    # ── Slovak ───────────────────────────────────────────────────────────────
+    Locale("sk", "Slovak", "Slovak", "sk-SK",
+           "Toto je náhľad hlasu zvukového popisu."),
+    # ── Swedish ──────────────────────────────────────────────────────────────
+    Locale("sv", "Swedish", "Swedish", "sv-SE",
+           "Det här är en förhandsgranskning av ljudbeskrivningsrösten."),
+]}
+
+# xlsx uses underscores; normalize to BCP-47 hyphen form
+_XLSX_ALIASES: dict[str, str] = {
+    code.replace("-", "_").lower(): code
+    for code in _REGISTRY
+    if "-" in code
+}
+# a few extra mappings for edge cases
+_XLSX_ALIASES.update({
+    "id": "id",      # Indonesian column header is just "id" (no region)
+})
+
+
+def normalize_code(code: str) -> str:
+    """
+    Normalize an arbitrary locale code to the canonical BCP-47 form used in this registry.
+
+    Handles:
+      - xlsx underscore form: "fr_fr" → "fr-FR"
+      - Bare language code: "fr" → "fr" (passthrough, legacy compat)
+      - Already canonical: "fr-FR" → "fr-FR"
+    """
+    if not code:
+        return code
+    lowered = code.strip().lower()
+    # e.g. "fr_fr" -> check alias table
+    if "_" in lowered:
+        return _XLSX_ALIASES.get(lowered, code.replace("_", "-").upper() if len(lowered) > 3 else code)
+    # Already hyphen form — canonicalise case
+    if "-" in code:
+        parts = code.split("-", 1)
+        canonical = f"{parts[0].lower()}-{parts[1].upper()}"
+        if canonical in _REGISTRY:
+            return canonical
+        return canonical
+    # Bare language code — return as-is (legacy)
+    return lowered
+
+
+def get(code: str) -> Locale | None:
+    """Return Locale for the given code, or None if unknown."""
+    canonical = normalize_code(code)
+    return _REGISTRY.get(canonical) or _REGISTRY.get(canonical.split("-")[0])
+
+
+def get_display_name(code: str) -> str:
+    """Human-readable display name, e.g. 'French (Canada)'."""
+    locale = get(code)
+    return locale.display_name if locale else code
+
+
+def get_gemini_label(code: str) -> str:
+    """
+    Label to use inside Gemini prompts, e.g. 'French (Canada)'.
+    Gemini models respond more reliably to human-readable language names
+    than to bare BCP-47 codes when used inside instruction prompts.
+    """
+    locale = get(code)
+    return locale.gemini_label if locale else code
+
+
+def get_tts_lang(code: str) -> str:
+    """BCP-47 code for the TTS API (may differ from canonical, e.g. es-MX → es-US)."""
+    locale = get(code)
+    return locale.tts_lang if locale else code
+
+
+def get_preview_sample(code: str) -> str:
+    """Language-appropriate TTS preview sentence."""
+    locale = get(code)
+    if locale:
+        return locale.preview_sample
+    # fallback: try parent language then English
+    parent = get(code.split("-")[0]) if "-" in code else None
+    if parent:
+        return parent.preview_sample
+    return "This is a preview of the audio description voice."
+
+
+def all_codes() -> list[str]:
+    """Return all registered locale codes, sorted."""
+    return sorted(_REGISTRY.keys())
+
+
+def all_display_map() -> dict[str, str]:
+    """Return {code: display_name} for all registered locales."""
+    return {code: locale.display_name for code, locale in _REGISTRY.items()}
diff --git a/backend/app/main.py b/backend/app/main.py
index 0b60f2c..86b8c79 100644
--- a/backend/app/main.py
+++ b/backend/app/main.py
@@ -23,6 +23,7 @@ from .api.v1.routes_tts import router as tts_router
 from .api.v1.routes_websockets import router as websockets_router
 from .api.v1.routes_vtt_versions import router as vtt_versions_router
 from .api.v1.routes_language_qc import router as language_qc_router
+from .api.v1.routes_glossaries import router as glossaries_router
 from .services.websocket import connection_manager
 from .core.config import settings
 from .core.secrets_config import initialize_config
@@ -273,6 +274,7 @@ app.include_router(jobs_router, prefix="/api/v1")
 app.include_router(review_notes_router, prefix="/api/v1")
 app.include_router(vtt_versions_router, prefix="/api/v1")
 app.include_router(language_qc_router, prefix="/api/v1")
+app.include_router(glossaries_router, prefix="/api/v1")
 app.include_router(tts_router, prefix="/api/v1")
 app.include_router(admin_router, prefix="/api/v1")
 app.include_router(websockets_router, prefix="/api/v1")
diff --git a/backend/app/models/audit_log.py b/backend/app/models/audit_log.py
index 1c96738..0a84153 100644
--- a/backend/app/models/audit_log.py
+++ b/backend/app/models/audit_log.py
@@ -61,6 +61,12 @@ class AuditAction(str, Enum):
     ADMIN_DATA_EXPORT = "admin.data.export"
     ADMIN_AUDIT_ACCESS = "admin.audit.access"
     
+    # Glossary management
+    GLOSSARY_UPLOAD = "glossary.upload"
+    GLOSSARY_VERSION_UPLOAD = "glossary.version.upload"
+    GLOSSARY_ACTIVATE = "glossary.activate"
+    GLOSSARY_ARCHIVE = "glossary.archive"
+
     # Security events
     RATE_LIMIT_EXCEEDED = "security.rate_limit.exceeded"
     VALIDATION_FAILURE = "security.validation.failure"
diff --git a/backend/app/models/glossary.py b/backend/app/models/glossary.py
new file mode 100644
index 0000000..0dfc60a
--- /dev/null
+++ b/backend/app/models/glossary.py
@@ -0,0 +1,139 @@
+from __future__ import annotations
+
+from datetime import datetime
+from enum import StrEnum
+
+from pydantic import BaseModel, Field
+
+
+class GlossarySource(StrEnum):
+    XLSX_UPLOAD = "xlsx_upload"
+    FRAZE_API = "fraze_api"  # reserved for future FRAZE integration
+
+
+class GlossaryStatus(StrEnum):
+    ACTIVE = "active"
+    ARCHIVED = "archived"
+
+
+class EmbeddingStatus(StrEnum):
+    PENDING = "pending"
+    IN_PROGRESS = "in_progress"
+    DONE = "done"
+    FAILED = "failed"
+
+
+class Glossary(BaseModel):
+    id: str | None = Field(None, alias="_id")
+    client_id: str
+    name: str
+    description: str | None = None
+    source_locale: str          # BCP-47 source column, e.g. "en-GB"
+    source: GlossarySource = GlossarySource.XLSX_UPLOAD
+    status: GlossaryStatus = GlossaryStatus.ACTIVE
+    current_version_id: str | None = None
+    created_at: datetime = Field(default_factory=datetime.utcnow)
+    created_by: str             # user_id
+
+    model_config = {"populate_by_name": True, "arbitrary_types_allowed": True}
+
+
+class GlossaryVersion(BaseModel):
+    id: str | None = Field(None, alias="_id")
+    glossary_id: str
+    version_number: int
+    source_xlsx_gcs_path: str | None = None  # GCS path to original file
+    term_count: int = 0
+    embedded_count: int = 0
+    embedding_status: EmbeddingStatus = EmbeddingStatus.PENDING
+    created_at: datetime = Field(default_factory=datetime.utcnow)
+    created_by: str
+    change_note: str | None = None
+
+    model_config = {"populate_by_name": True}
+
+
+class GlossaryTerm(BaseModel):
+    """One source term with its per-locale translations."""
+    id: str | None = Field(None, alias="_id")
+    glossary_id: str
+    version_id: str
+    cid: str | None = None   # 3M Content ID from xlsx
+    tid: str | None = None   # 3M Term ID from xlsx
+    source_term: str            # canonical source text (whitespace-normalised)
+    source_term_lower: str      # lowercase for case-insensitive index
+    translations: dict[str, str] = {}  # {locale_code: translated_text}
+    embedding: list[float] | None = None  # 768-dim Gemini embedding
+
+    model_config = {"populate_by_name": True}
+
+
+# ── Schema models (API request/response) ──────────────────────────────────────
+
+class GlossaryCreate(BaseModel):
+    name: str
+    description: str | None = None
+    source_locale: str
+    change_note: str | None = None
+
+
+class GlossaryVersionCreate(BaseModel):
+    source_locale: str
+    change_note: str | None = None
+
+
+class GlossaryResponse(BaseModel):
+    id: str
+    client_id: str
+    name: str
+    description: str | None = None
+    source_locale: str
+    source: GlossarySource
+    status: GlossaryStatus
+    current_version_id: str | None = None
+    created_at: datetime
+    created_by: str
+
+
+class GlossaryVersionResponse(BaseModel):
+    id: str
+    glossary_id: str
+    version_number: int
+    term_count: int
+    embedded_count: int
+    embedding_status: EmbeddingStatus
+    created_at: datetime
+    created_by: str
+    change_note: str | None = None
+
+
+class GlossaryDetailResponse(GlossaryResponse):
+    versions: list[GlossaryVersionResponse] = []
+
+
+class GlossaryTermPreview(BaseModel):
+    """Subset of GlossaryTerm for UI previews."""
+    source_term: str
+    translations: dict[str, str]
+
+
+class MatchedTerm(BaseModel):
+    """A term matched against VTT source text, with the target-locale translation."""
+    source_term: str
+    target_translation: str
+    match_kind: str  # "exact" | "vector"
+    score: float     # 1.0 for exact, cosine similarity for vector
+
+
+def glossary_from_doc(doc: dict) -> Glossary:
+    doc = dict(doc)
+    if "_id" in doc:
+        doc["_id"] = str(doc["_id"])
+    return Glossary.model_validate(doc)
+
+
+def glossary_version_from_doc(doc: dict) -> GlossaryVersion:
+    doc = dict(doc)
+    if "_id" in doc:
+        doc["_id"] = str(doc["_id"])
+    return GlossaryVersion.model_validate(doc)
diff --git a/backend/app/prompts/gemini_ingestion.md b/backend/app/prompts/gemini_ingestion.md
index f52a745..b471d66 100644
--- a/backend/app/prompts/gemini_ingestion.md
+++ b/backend/app/prompts/gemini_ingestion.md
@@ -47,6 +47,8 @@ BRAND NAMES AND PRODUCTS:
 - If a product is on the brand list, use the brand name even if the label is partially obscured — use your best confident identification
 - If a product is NOT on the list or is completely unclear, use a generic descriptor — do not invent brand names
 
+{GLOSSARY}
+
 CAPTION FORMATTING (DCMP standard):
 - Maximum TWO lines per caption. Never use three or more lines.
 - Each line should be no longer than ~37 characters where possible (42 absolute max)
diff --git a/backend/app/prompts/gemini_ingestion_targeted.md b/backend/app/prompts/gemini_ingestion_targeted.md
index 4af276b..f9f0efd 100644
--- a/backend/app/prompts/gemini_ingestion_targeted.md
+++ b/backend/app/prompts/gemini_ingestion_targeted.md
@@ -51,6 +51,8 @@ BRAND NAMES AND PRODUCTS:
 - If a product is on the brand list, use the brand name even if the label is partially obscured — use your best confident identification
 - If a product is NOT on the list or is completely unclear, use a generic descriptor — do not invent brand names
 
+{GLOSSARY}
+
 CAPTION FORMATTING (DCMP standard):
 - Maximum TWO lines per caption. Never use three or more lines.
 - Each line should be no longer than ~37 characters where possible (42 absolute max)
diff --git a/backend/app/prompts/gemini_transcreation.md b/backend/app/prompts/gemini_transcreation.md
index 13f38d4..0255cc1 100644
--- a/backend/app/prompts/gemini_transcreation.md
+++ b/backend/app/prompts/gemini_transcreation.md
@@ -7,6 +7,8 @@ Rewrite the following English captions and audio descriptions into {TARGET_LANGU
 - timing boundaries (same cue timestamps),
 - line lengths friendly for readability (~32–40 chars).
 
+{GLOSSARY}
+
 Input:
 - captions_vtt_en: <VTT text>
 - ad_vtt_en: <VTT text>
diff --git a/backend/app/services/embedding_service.py b/backend/app/services/embedding_service.py
new file mode 100644
index 0000000..3c703a7
--- /dev/null
+++ b/backend/app/services/embedding_service.py
@@ -0,0 +1,72 @@
+"""
+Embedding service backed by Gemini text-embedding-004.
+
+Provides batch embedding with retry/backoff for use in glossary ingestion.
+Batch size: 100 texts per API call (API limit is 2048 but we keep it conservative
+for memory and retry ergonomics with large glossaries).
+"""
+from __future__ import annotations
+
+import asyncio
+from collections.abc import Sequence
+
+from google import genai
+from google.genai import types as genai_types
+
+from ..core.config import settings
+from ..core.logging import get_logger
+
+logger = get_logger(__name__)
+
+_EMBED_MODEL = "text-embedding-004"
+_BATCH_SIZE = 100
+_MAX_RETRIES = 3
+_INITIAL_BACKOFF = 2.0
+
+
+class EmbeddingService:
+    def __init__(self) -> None:
+        self._client = genai.Client(api_key=settings.gemini_api_key)
+
+    async def embed_texts(self, texts: Sequence[str]) -> list[list[float]]:
+        """
+        Embed a list of texts and return a list of 768-dim float vectors.
+        Processes in batches; retries with exponential backoff on transient errors.
+        Order is preserved.
+        """
+        results: list[list[float]] = []
+        for i in range(0, len(texts), _BATCH_SIZE):
+            batch = list(texts[i: i + _BATCH_SIZE])
+            vectors = await self._embed_batch_with_retry(batch)
+            results.extend(vectors)
+        return results
+
+    async def embed_text(self, text: str) -> list[float]:
+        vectors = await self.embed_texts([text])
+        return vectors[0]
+
+    async def _embed_batch_with_retry(self, texts: list[str]) -> list[list[float]]:
+        backoff = _INITIAL_BACKOFF
+        for attempt in range(1, _MAX_RETRIES + 1):
+            try:
+                response = await asyncio.to_thread(
+                    self._client.models.embed_content,
+                    model=_EMBED_MODEL,
+                    contents=texts,
+                    config=genai_types.EmbedContentConfig(
+                        task_type="RETRIEVAL_DOCUMENT",
+                    ),
+                )
+                return [list(emb.values) for emb in response.embeddings]
+            except Exception as exc:
+                if attempt == _MAX_RETRIES:
+                    logger.error(f"Embedding batch failed after {_MAX_RETRIES} attempts: {exc}")
+                    raise
+                logger.warning(f"Embedding attempt {attempt} failed, retrying in {backoff}s: {exc}")
+                await asyncio.sleep(backoff)
+                backoff *= 2
+
+        raise RuntimeError("unreachable")  # makes type-checker happy
+
+
+embedding_service = EmbeddingService()
diff --git a/backend/app/services/gemini.py b/backend/app/services/gemini.py
index 72b351e..b281175 100644
--- a/backend/app/services/gemini.py
+++ b/backend/app/services/gemini.py
@@ -8,6 +8,7 @@ import google.genai as genai
 
 from ..core.config import settings
 from ..core.logging import get_logger
+from ..lib import locales as locale_lib
 
 logger = get_logger(__name__)
 
@@ -106,6 +107,12 @@ Generate sdh_captions_vtt using the same cue timings as captions_vtt, enriched w
 - Maintain the same timestamp format as captions_vtt (HH:MM:SS.mmm --> HH:MM:SS.mmm)
 - Only add sound effect cues where they add meaningful context; do not annotate every minor sound"""
 
+    def _build_glossary_block(self, glossary_block: Optional[str]) -> str:
+        """Return the pre-built glossary block (from glossary_service.build_glossary_prompt_block), or empty string."""
+        if glossary_block and glossary_block.strip():
+            return glossary_block.strip()
+        return ""
+
     def _build_brand_context_block(self, brand_context: Optional[str]) -> str:
         """Build the brand context instruction block for injection into prompts."""
         if brand_context and brand_context.strip():
@@ -118,7 +125,7 @@ Generate sdh_captions_vtt using the same cue timings as captions_vtt, enriched w
                 )
         return "No specific brand names have been provided for this video."
 
-    async def extract_accessibility(self, video_file_path: str, brand_context: Optional[str] = None, sdh_requested: bool = False, _cost_ctx: Optional[dict] = None) -> dict[str, Any]:
+    async def extract_accessibility(self, video_file_path: str, brand_context: Optional[str] = None, sdh_requested: bool = False, glossary_block: Optional[str] = None, _cost_ctx: Optional[dict] = None) -> dict[str, Any]:
         """
         Extract captions and audio descriptions from video using Gemini 2.0
         Returns structured JSON with transcript, captions VTT, and audio description VTT
@@ -127,6 +134,7 @@ Generate sdh_captions_vtt using the same cue timings as captions_vtt, enriched w
         prompt = (
             prompt_template
             .replace("{BRAND_CONTEXT}", self._build_brand_context_block(brand_context))
+            .replace("{GLOSSARY}", self._build_glossary_block(glossary_block))
             .replace("{SDH_FIELD}", self._build_sdh_field(sdh_requested))
             .replace("{SDH_GUIDELINES}", self._build_sdh_guidelines(sdh_requested))
         )
@@ -320,6 +328,7 @@ Fix the JSON and return it:
         target_language: str,
         brand_context: Optional[str] = None,
         sdh_requested: bool = False,
+        glossary_block: Optional[str] = None,
         _cost_ctx: Optional[dict] = None,
     ) -> dict[str, Any]:
         """
@@ -343,8 +352,9 @@ Fix the JSON and return it:
         prompt_template = self._load_prompt("gemini_ingestion_targeted.md")
         prompt = (
             prompt_template
-            .replace("{TARGET_LANGUAGE}", target_language)
+            .replace("{TARGET_LANGUAGE}", locale_lib.get_gemini_label(target_language))
             .replace("{BRAND_CONTEXT}", self._build_brand_context_block(brand_context))
+            .replace("{GLOSSARY}", self._build_glossary_block(glossary_block))
             .replace("{SDH_FIELD}", self._build_sdh_field(sdh_requested))
             .replace("{SDH_GUIDELINES}", self._build_sdh_guidelines(sdh_requested))
         )
@@ -756,6 +766,7 @@ Fix the JSON and return it:
         ad_vtt: str,
         target_language: str,
         brief: Optional[str] = None,
+        glossary_block: Optional[str] = None,
         _cost_ctx: Optional[dict] = None,
     ) -> dict[str, str]:
         """
@@ -765,7 +776,8 @@ Fix the JSON and return it:
 
         # Format prompt with actual content
         prompt = prompt_template.format(
-            TARGET_LANGUAGE=target_language
+            TARGET_LANGUAGE=locale_lib.get_gemini_label(target_language),
+            GLOSSARY=self._build_glossary_block(glossary_block),
         )
 
         user_prompt = f"""
@@ -817,6 +829,7 @@ JSON:
         vtt_content: str,
         target_language: str,
         source_language: str = "en",
+        glossary_block: Optional[str] = None,
         _cost_ctx: Optional[dict] = None,
     ) -> str:
         """
@@ -842,14 +855,18 @@ JSON:
                 f"{i + 1}. {cue.text.replace(chr(10), ' ')}"
                 for i, cue in enumerate(source_cues)
             )
-            prompt = f"""Translate the following {cue_count} numbered text segments from {source_language} to {target_language}.
+            _src_label = locale_lib.get_gemini_label(source_language)
+            _tgt_label = locale_lib.get_gemini_label(target_language)
+            _glossary_section = self._build_glossary_block(glossary_block)
+            _glossary_line = f"\n\n{_glossary_section}" if _glossary_section else ""
+            prompt = f"""Translate the following {cue_count} numbered text segments from {_src_label} to {_tgt_label}.
 
 REQUIREMENTS:
 - Return EXACTLY {cue_count} numbered lines, one translation per line
 - Format: "1. translated text", "2. translated text", etc.
 - Preserve speaker labels like [Speaker 1]: unchanged
-- Use natural, idiomatic {target_language}
-- Do NOT add any explanation, preamble, or extra lines{extra_instruction}
+- Use natural, idiomatic {_tgt_label}
+- Do NOT add any explanation, preamble, or extra lines{extra_instruction}{_glossary_line}
 
 Segments to translate:
 {numbered_texts}"""
diff --git a/backend/app/services/gemini_tts.py b/backend/app/services/gemini_tts.py
index 3a2a1a6..c559866 100644
--- a/backend/app/services/gemini_tts.py
+++ b/backend/app/services/gemini_tts.py
@@ -7,6 +7,7 @@ from pydub import AudioSegment
 
 from ..core.config import settings
 from ..core.logging import get_logger
+from ..lib import locales as locale_lib
 
 logger = get_logger(__name__)
 
@@ -166,10 +167,10 @@ class GeminiTTSService:
         Generate a preview audio sample for voice selection.
         Uses language-specific sample text and applies all TTS settings.
         """
-        # Get preview sample text for the language
-        sample_text = settings.gemini_tts_preview_samples.get(
-            language,
-            settings.gemini_tts_preview_samples.get("en", "This is a voice preview.")
+        # Get preview sample text — try settings override, then locale registry, then fallback
+        sample_text = (
+            settings.gemini_tts_preview_samples.get(language)
+            or locale_lib.get_preview_sample(language)
         )
 
         return await self.synthesize_text(
diff --git a/backend/app/services/glossary_service.py b/backend/app/services/glossary_service.py
new file mode 100644
index 0000000..5c96742
--- /dev/null
+++ b/backend/app/services/glossary_service.py
@@ -0,0 +1,736 @@
+"""
+Glossary service — per-client terminology management.
+
+Responsibilities:
+  • parse_xlsx(bytes, source_col) → list of (source_term, {locale: translation})
+  • ingest_glossary(...)          → create Glossary + GlossaryVersion + GlossaryTerms in Mongo
+  • activate_version(...)         → atomic swap of current_version_id
+  • match_terms_for_text(...)     → hybrid exact + vector retrieval for prompt injection
+  • build_glossary_prompt_block(...)  → formats matched terms for the Gemini prompt
+"""
+from __future__ import annotations
+
+import io
+import re
+from collections.abc import Sequence
+from dataclasses import dataclass
+from datetime import datetime
+
+from bson import ObjectId
+from fastapi import UploadFile
+
+from ..core.database import get_database
+from ..core.logging import get_logger
+from ..lib import locales as locale_lib
+from ..models.glossary import (
+    EmbeddingStatus,
+    Glossary,
+    GlossaryStatus,
+    GlossaryTerm,
+    GlossaryVersion,
+    MatchedTerm,
+    glossary_from_doc,
+    glossary_version_from_doc,
+)
+
+logger = get_logger(__name__)
+
+_COLL_GLOSSARIES = "glossaries"
+_COLL_VERSIONS = "glossary_versions"
+_COLL_TERMS = "glossary_terms"
+
+# Maximum number of terms injected into a single Gemini prompt
+_MAX_TERMS_IN_PROMPT = 50
+
+# Atlas Vector Search index name (must exist on the collection)
+_VECTOR_INDEX = "glossary_embedding_index"
+_VECTOR_DIMS = 768
+_VECTOR_SIMILARITY_THRESHOLD = 0.75
+_VECTOR_TOP_K = 20
+
+
+# ── xlsx parsing ─────────────────────────────────────────────────────────────
+
+@dataclass
+class _ParsedTerm:
+    cid: str | None
+    tid: str | None
+    source_term: str
+    translations: dict[str, str]  # {normalized_locale: text}
+
+
+def _cell(row: tuple, idx: int | None) -> str | None:
+    if idx is None or idx >= len(row):
+        return None
+    v = row[idx]
+    return str(v).strip() if v is not None else None
+
+
+def parse_xlsx(file_bytes: bytes, source_locale_col: str) -> list[_ParsedTerm]:
+    """
+    Parse an xlsx glossary file.
+
+    Args:
+        file_bytes: Raw xlsx bytes.
+        source_locale_col: The column header that contains the source text,
+                           e.g. "en_gb" or "en-GB". Case-insensitive.
+
+    Returns:
+        List of parsed terms. Rows where the source column is empty are skipped.
+    """
+    import openpyxl  # local import — only used during ingest
+
+    wb = openpyxl.load_workbook(io.BytesIO(file_bytes), read_only=True, data_only=True)
+    ws = wb.active
+
+    rows = ws.iter_rows(values_only=True)
+    try:
+        header_row = next(rows)
+    except StopIteration:
+        return []
+
+    # Normalise header names to canonical locale codes
+    headers: list[str | None] = []
+    for h in header_row:
+        if h is None:
+            headers.append(None)
+            continue
+        s = str(h).strip()
+        headers.append(s)
+
+    # Find column indices
+    src_col_name = source_locale_col.strip()
+    # Try exact match first, then case-insensitive
+    src_idx: int | None = None
+    for i, h in enumerate(headers):
+        if h and h.lower() == src_col_name.lower():
+            src_idx = i
+            break
+
+    if src_idx is None:
+        raise ValueError(f"Source column '{source_locale_col}' not found in xlsx. Available: {[h for h in headers if h]}")
+
+    cid_idx = next((i for i, h in enumerate(headers) if h and h.upper() == "CID"), None)
+    tid_idx = next((i for i, h in enumerate(headers) if h and h.upper() == "TID"), None)
+
+    # All other columns with valid locale-like names become translation columns
+    locale_cols: list[tuple[int, str]] = []  # [(col_index, normalized_locale_code)]
+    for i, h in enumerate(headers):
+        if h is None or i == src_idx or i == cid_idx or i == tid_idx:
+            continue
+        norm = locale_lib.normalize_code(h)
+        if norm:
+            locale_cols.append((i, norm))
+
+    terms: list[_ParsedTerm] = []
+    for row in rows:
+        if not row or all(v is None for v in row):
+            continue
+
+        source = _cell(row, src_idx)
+        if not source:
+            continue
+
+        translations: dict[str, str] = {}
+        for col_idx, locale_code in locale_cols:
+            val = _cell(row, col_idx)
+            if val:
+                translations[locale_code] = val
+
+        terms.append(_ParsedTerm(
+            cid=_cell(row, cid_idx),
+            tid=_cell(row, tid_idx),
+            source_term=source,
+            translations=translations,
+        ))
+
+    wb.close()
+    return terms
+
+
+# ── Ingest ────────────────────────────────────────────────────────────────────
+
+async def ingest_glossary(
+    client_id: str,
+    name: str,
+    source_locale: str,
+    source_locale_col: str,
+    file: UploadFile,
+    user_id: str,
+    description: str | None = None,
+    change_note: str | None = None,
+) -> tuple[Glossary, GlossaryVersion]:
+    """
+    Full glossary ingestion pipeline:
+      1. Upload xlsx to GCS
+      2. Parse terms
+      3. Create Glossary + GlossaryVersion + GlossaryTerm documents in Mongo
+      4. Kick off background embedding task
+
+    Returns (Glossary, GlossaryVersion) on success.
+    """
+    db = await get_database()
+
+    # ── Upload original xlsx to GCS ──
+    file_bytes = await file.read()
+    glossary_id = str(ObjectId())
+    version_id = str(ObjectId())
+    gcs_path = f"glossaries/{client_id}/{glossary_id}/{version_id}/source.xlsx"
+    await _upload_bytes_to_gcs(file_bytes, gcs_path,
+                               content_type="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")
+
+    # ── Parse ──
+    logger.info(f"Parsing xlsx for glossary {glossary_id}, source_col={source_locale_col}")
+    parsed_terms = parse_xlsx(file_bytes, source_locale_col)
+    logger.info(f"Parsed {len(parsed_terms)} terms")
+
+    # ── Create Glossary doc ──
+    now = datetime.utcnow()
+    glossary_doc = {
+        "_id": ObjectId(glossary_id),
+        "client_id": client_id,
+        "name": name,
+        "description": description,
+        "source_locale": locale_lib.normalize_code(source_locale),
+        "source": "xlsx_upload",
+        "status": GlossaryStatus.ACTIVE.value,
+        "current_version_id": version_id,
+        "created_at": now,
+        "created_by": user_id,
+    }
+    await db[_COLL_GLOSSARIES].insert_one(glossary_doc)
+
+    # ── Create GlossaryVersion doc ──
+    version_doc = {
+        "_id": ObjectId(version_id),
+        "glossary_id": glossary_id,
+        "version_number": 1,
+        "source_xlsx_gcs_path": gcs_path,
+        "term_count": len(parsed_terms),
+        "embedded_count": 0,
+        "embedding_status": EmbeddingStatus.PENDING.value,
+        "created_at": now,
+        "created_by": user_id,
+        "change_note": change_note,
+    }
+    await db[_COLL_VERSIONS].insert_one(version_doc)
+
+    # ── Bulk insert GlossaryTerms ──
+    if parsed_terms:
+        term_docs = [
+            {
+                "_id": ObjectId(),
+                "glossary_id": glossary_id,
+                "version_id": version_id,
+                "cid": t.cid,
+                "tid": t.tid,
+                "source_term": t.source_term,
+                "source_term_lower": t.source_term.lower(),
+                "translations": t.translations,
+                "embedding": None,
+            }
+            for t in parsed_terms
+        ]
+        await db[_COLL_TERMS].insert_many(term_docs, ordered=False)
+
+    # ── Create collection indexes (idempotent) ──
+    await _ensure_indexes(db)
+
+    # ── Kick off embedding Celery task ──
+    try:
+        from ..tasks.embed_glossary import embed_glossary_version_task
+        embed_glossary_version_task.delay(version_id)
+        logger.info(f"Queued embedding task for version {version_id}")
+    except Exception as e:
+        logger.warning(f"Could not queue embedding task: {e}")
+
+    glossary = glossary_from_doc(glossary_doc)
+    version = glossary_version_from_doc(version_doc)
+    return glossary, version
+
+
+async def ingest_new_version(
+    glossary_id: str,
+    source_locale_col: str,
+    file: UploadFile,
+    user_id: str,
+    change_note: str | None = None,
+) -> GlossaryVersion:
+    """Add a new version to an existing glossary without replacing it as active."""
+    db = await get_database()
+
+    glossary_doc = await db[_COLL_GLOSSARIES].find_one({"_id": ObjectId(glossary_id)})
+    if not glossary_doc:
+        raise ValueError(f"Glossary {glossary_id} not found")
+
+    client_id = glossary_doc["client_id"]
+
+    # Find next version number
+    last_version = await db[_COLL_VERSIONS].find_one(
+        {"glossary_id": glossary_id},
+        sort=[("version_number", -1)],
+    )
+    next_version_num = (last_version["version_number"] + 1) if last_version else 1
+
+    file_bytes = await file.read()
+    version_id = str(ObjectId())
+    gcs_path = f"glossaries/{client_id}/{glossary_id}/{version_id}/source.xlsx"
+    await _upload_bytes_to_gcs(file_bytes, gcs_path,
+                               content_type="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet")
+
+    parsed_terms = parse_xlsx(file_bytes, source_locale_col)
+
+    now = datetime.utcnow()
+    version_doc = {
+        "_id": ObjectId(version_id),
+        "glossary_id": glossary_id,
+        "version_number": next_version_num,
+        "source_xlsx_gcs_path": gcs_path,
+        "term_count": len(parsed_terms),
+        "embedded_count": 0,
+        "embedding_status": EmbeddingStatus.PENDING.value,
+        "created_at": now,
+        "created_by": user_id,
+        "change_note": change_note,
+    }
+    await db[_COLL_VERSIONS].insert_one(version_doc)
+
+    if parsed_terms:
+        term_docs = [
+            {
+                "_id": ObjectId(),
+                "glossary_id": glossary_id,
+                "version_id": version_id,
+                "cid": t.cid,
+                "tid": t.tid,
+                "source_term": t.source_term,
+                "source_term_lower": t.source_term.lower(),
+                "translations": t.translations,
+                "embedding": None,
+            }
+            for t in parsed_terms
+        ]
+        await db[_COLL_TERMS].insert_many(term_docs, ordered=False)
+
+    try:
+        from ..tasks.embed_glossary import embed_glossary_version_task
+        embed_glossary_version_task.delay(version_id)
+    except Exception as e:
+        logger.warning(f"Could not queue embedding task: {e}")
+
+    return glossary_version_from_doc(version_doc)
+
+
+async def activate_version(glossary_id: str, version_id: str) -> None:
+    """Atomically set the active version of a glossary."""
+    db = await get_database()
+    result = await db[_COLL_GLOSSARIES].update_one(
+        {"_id": ObjectId(glossary_id)},
+        {"$set": {"current_version_id": version_id}},
+    )
+    if result.matched_count == 0:
+        raise ValueError(f"Glossary {glossary_id} not found")
+    # Invalidate Redis cache
+    await _invalidate_cache(glossary_id)
+
+
+async def archive_glossary(glossary_id: str) -> None:
+    db = await get_database()
+    await db[_COLL_GLOSSARIES].update_one(
+        {"_id": ObjectId(glossary_id)},
+        {"$set": {"status": GlossaryStatus.ARCHIVED.value}},
+    )
+    await _invalidate_cache(glossary_id)
+
+
+# ── Retrieval ─────────────────────────────────────────────────────────────────
+
+async def match_terms_for_text(
+    client_id: str,
+    text: str,
+    target_locale: str,
+    top_k: int = _MAX_TERMS_IN_PROMPT,
+) -> list[MatchedTerm]:
+    """
+    Hybrid retrieval: exact-match (Aho-Corasick) + semantic (Atlas Vector Search).
+
+    Returns a ranked, deduplicated list of up to `top_k` MatchedTerm objects,
+    each with the source term and its translation in `target_locale`.
+    Exact matches rank before vector matches.
+    """
+    db = await get_database()
+    norm_target = locale_lib.normalize_code(target_locale)
+
+    active_version_id = await _get_active_version_id(client_id)
+    if not active_version_id:
+        return []
+
+    # ── Exact pass ──
+    exact_matches = await _exact_match(db, active_version_id, text, norm_target)
+
+    # ── Vector pass (if we haven't hit the limit yet) ──
+    remaining = top_k - len(exact_matches)
+    already_found = {m.source_term.lower() for m in exact_matches}
+    vector_matches: list[MatchedTerm] = []
+
+    if remaining > 0:
+        try:
+            vector_matches = await _vector_match(
+                db, active_version_id, text, norm_target,
+                top_k=_VECTOR_TOP_K, exclude_terms=already_found,
+            )
+        except Exception as e:
+            logger.warning(f"Vector search failed (non-fatal): {e}")
+
+    combined = exact_matches + vector_matches
+    if len(combined) > top_k:
+        logger.info(f"glossary_terms_truncated: had {len(combined)}, capped at {top_k}")
+        combined = combined[:top_k]
+
+    return combined
+
+
+async def _get_active_version_id(client_id: str) -> str | None:
+    """Return the active version_id for the active glossary of a client, or None."""
+    try:
+        from ..core.redis import redis_client  # lazy import
+        cache_key = f"glossary:active_version:{client_id}"
+        cached = await redis_client.get(cache_key)
+        if cached:
+            return cached.decode() if isinstance(cached, bytes) else cached
+    except Exception:
+        pass
+
+    db = await get_database()
+    glossary_doc = await db[_COLL_GLOSSARIES].find_one(
+        {"client_id": client_id, "status": GlossaryStatus.ACTIVE.value},
+        sort=[("created_at", -1)],
+    )
+    if not glossary_doc or not glossary_doc.get("current_version_id"):
+        return None
+
+    version_id = glossary_doc["current_version_id"]
+
+    try:
+        from ..core.redis import redis_client
+        cache_key = f"glossary:active_version:{client_id}"
+        await redis_client.setex(cache_key, 3600, version_id)
+    except Exception:
+        pass
+
+    return version_id
+
+
+async def _invalidate_cache(glossary_id: str) -> None:
+    """Clear Redis cache for a glossary's client."""
+    try:
+        db = await get_database()
+        doc = await db[_COLL_GLOSSARIES].find_one({"_id": ObjectId(glossary_id)})
+        if doc:
+            from ..core.redis import redis_client
+            await redis_client.delete(f"glossary:active_version:{doc['client_id']}")
+    except Exception as e:
+        logger.debug(f"Cache invalidation skipped: {e}")
+
+
+async def _exact_match(
+    db,
+    version_id: str,
+    text: str,
+    target_locale: str,
+) -> list[MatchedTerm]:
+    """Find terms present in `text` using Aho-Corasick over the glossary terms."""
+    import ahocorasick  # pyahocorasick
+
+    # Load all terms for this version (source_term_lower + translations)
+    cursor = db[_COLL_TERMS].find(
+        {"version_id": version_id},
+        {"source_term": 1, "source_term_lower": 1, "translations": 1},
+    )
+    terms = await cursor.to_list(length=None)
+    if not terms:
+        return []
+
+    # Build automaton
+    automaton = ahocorasick.Automaton()
+    for doc in terms:
+        stl = doc["source_term_lower"]
+        automaton.add_word(stl, (doc["source_term"], doc["translations"]))
+    automaton.make_automaton()
+
+    text_lower = text.lower()
+    matched: list[MatchedTerm] = []
+    seen: set[str] = set()
+
+    for _end_idx, (source_term, translations) in automaton.iter(text_lower):
+        if source_term in seen:
+            continue
+        # Require word/phrase boundaries around the match
+        start_idx = _end_idx - len(source_term.lower()) + 1
+        if start_idx > 0 and text_lower[start_idx - 1].isalnum():
+            continue
+        end_after = _end_idx + 1
+        if end_after < len(text_lower) and text_lower[end_after].isalnum():
+            continue
+
+        target_text = _get_translation(translations, target_locale)
+        if not target_text:
+            continue
+        seen.add(source_term)
+        matched.append(MatchedTerm(
+            source_term=source_term,
+            target_translation=target_text,
+            match_kind="exact",
+            score=1.0,
+        ))
+
+    return matched
+
+
+async def _vector_match(
+    db,
+    version_id: str,
+    text: str,
+    target_locale: str,
+    top_k: int = 20,
+    exclude_terms: set[str] | None = None,
+) -> list[MatchedTerm]:
+    """Semantic search via Atlas Vector Search ($vectorSearch)."""
+    from ..services.embedding_service import embedding_service
+
+    query_embedding = await embedding_service.embed_text(text[:2000])  # cap input length
+
+    pipeline = [
+        {
+            "$vectorSearch": {
+                "index": _VECTOR_INDEX,
+                "path": "embedding",
+                "queryVector": query_embedding,
+                "numCandidates": top_k * 4,
+                "limit": top_k,
+                "filter": {"version_id": version_id},
+            }
+        },
+        {
+            "$project": {
+                "source_term": 1,
+                "translations": 1,
+                "score": {"$meta": "vectorSearchScore"},
+            }
+        },
+    ]
+
+    cursor = db[_COLL_TERMS].aggregate(pipeline)
+    results = await cursor.to_list(length=top_k)
+
+    matched: list[MatchedTerm] = []
+    for doc in results:
+        score = doc.get("score", 0.0)
+        if score < _VECTOR_SIMILARITY_THRESHOLD:
+            continue
+        source_term = doc["source_term"]
+        if exclude_terms and source_term.lower() in exclude_terms:
+            continue
+        target_text = _get_translation(doc["translations"], target_locale)
+        if not target_text:
+            continue
+        matched.append(MatchedTerm(
+            source_term=source_term,
+            target_translation=target_text,
+            match_kind="vector",
+            score=score,
+        ))
+
+    return matched
+
+
+def _get_translation(translations: dict[str, str], target_locale: str) -> str | None:
+    """Look up a translation with locale-fallback: fr-CA → fr-FR → fr → None."""
+    if not translations:
+        return None
+    if target_locale in translations:
+        return translations[target_locale]
+    # Try parent language
+    parent = target_locale.split("-")[0] if "-" in target_locale else None
+    if parent:
+        # Try sibling locales, e.g. fr-CA not found → try fr-FR
+        for code, text in translations.items():
+            if code.startswith(parent + "-") or code == parent:
+                return text
+    return None
+
+
+# ── Prompt block ──────────────────────────────────────────────────────────────
+
+def build_glossary_prompt_block(
+    matched_terms: Sequence[MatchedTerm],
+    target_locale: str,
+) -> str:
+    """
+    Format matched terms for injection into a Gemini prompt.
+    Returns an empty string if no terms were matched.
+    """
+    if not matched_terms:
+        return ""
+
+    target_label = locale_lib.get_gemini_label(target_locale)
+    lines = [
+        f"## Approved {target_label} terminology",
+        "Use these exact translations when the source terms appear — do not deviate:",
+    ]
+    for term in matched_terms:
+        lines.append(f'- "{term.source_term}" → "{term.target_translation}"')
+
+    return "\n".join(lines)
+
+
+# ── Helpers ───────────────────────────────────────────────────────────────────
+
+async def _upload_bytes_to_gcs(data: bytes, gcs_path: str, content_type: str) -> None:
+    import asyncio
+    loop = asyncio.get_event_loop()
+
+    def _upload() -> None:
+        from google.cloud import storage as gcs_storage
+
+        from ..core.config import settings
+        client = gcs_storage.Client(project=settings.gcp_project_id)
+        bucket = client.bucket(settings.gcs_bucket)
+        blob = bucket.blob(gcs_path)
+        blob.content_type = content_type
+        blob.upload_from_string(data, content_type=content_type)
+
+    await loop.run_in_executor(None, _upload)
+
+
+async def _ensure_indexes(db) -> None:
+    try:
+        await db[_COLL_GLOSSARIES].create_index([("client_id", 1), ("status", 1)])
+        await db[_COLL_VERSIONS].create_index([("glossary_id", 1), ("version_number", -1)])
+        await db[_COLL_TERMS].create_index([("version_id", 1), ("source_term_lower", 1)])
+        await db[_COLL_TERMS].create_index([("glossary_id", 1)])
+    except Exception as e:
+        logger.debug(f"Index creation skipped (likely already exist): {e}")
+
+
+# ── Task helpers ─────────────────────────────────────────────────────────────
+
+async def get_glossary_block_for_job(
+    job_doc: dict,
+    target_locale: str,
+    db,
+) -> str:
+    """
+    Convenience function for Celery tasks: given a job document and a target locale,
+    return the formatted glossary block for Gemini prompt injection (or empty string).
+
+    Looks up:
+      job_doc.project_id → db.projects → client_id → active glossary version
+
+    Non-fatal: any failure returns "" so the pipeline continues without a glossary.
+    """
+    try:
+        project_id = job_doc.get("project_id")
+        if not project_id:
+            return ""
+
+        project = await db.projects.find_one({"_id": project_id})
+        if not project:
+            return ""
+
+        client_id = project.get("client_id")
+        if not client_id:
+            return ""
+
+        # Get active version id via our cache-backed helper (reuses Redis if available)
+        active_version_id = await _get_active_version_id(client_id)
+        if not active_version_id:
+            return ""
+
+        # Combine source VTT texts for matching
+        source_text = job_doc.get("_glossary_source_text", "")
+        if not source_text:
+            return ""
+
+        norm_target = locale_lib.normalize_code(target_locale)
+        exact_matches = await _exact_match(db, active_version_id, source_text, norm_target)
+
+        remaining = _MAX_TERMS_IN_PROMPT - len(exact_matches)
+        already_found = {m.source_term.lower() for m in exact_matches}
+        vector_matches: list[MatchedTerm] = []
+
+        if remaining > 0:
+            try:
+                vector_matches = await _vector_match(
+                    db, active_version_id, source_text, norm_target,
+                    top_k=_VECTOR_TOP_K, exclude_terms=already_found,
+                )
+            except Exception as ve:
+                logger.debug(f"Vector search skipped in task context: {ve}")
+
+        combined = exact_matches + vector_matches
+        if len(combined) > _MAX_TERMS_IN_PROMPT:
+            logger.info(f"glossary_terms_truncated: capped at {_MAX_TERMS_IN_PROMPT}")
+            combined = combined[:_MAX_TERMS_IN_PROMPT]
+
+        return build_glossary_prompt_block(combined, target_locale)
+
+    except Exception as e:
+        logger.warning(f"Glossary lookup failed for job {job_doc.get('_id')} (non-fatal): {e}")
+        return ""
+
+
+# ── Listing helpers ───────────────────────────────────────────────────────────
+
+async def get_glossaries_for_client(client_id: str) -> list[Glossary]:
+    db = await get_database()
+    cursor = db[_COLL_GLOSSARIES].find(
+        {"client_id": client_id, "status": {"$ne": GlossaryStatus.ARCHIVED.value}},
+        sort=[("created_at", -1)],
+    )
+    docs = await cursor.to_list(length=100)
+    return [glossary_from_doc(d) for d in docs]
+
+
+async def get_glossary(glossary_id: str) -> Glossary | None:
+    db = await get_database()
+    doc = await db[_COLL_GLOSSARIES].find_one({"_id": ObjectId(glossary_id)})
+    return glossary_from_doc(doc) if doc else None
+
+
+async def get_versions(glossary_id: str) -> list[GlossaryVersion]:
+    db = await get_database()
+    cursor = db[_COLL_VERSIONS].find(
+        {"glossary_id": glossary_id},
+        sort=[("version_number", -1)],
+    )
+    docs = await cursor.to_list(length=50)
+    return [glossary_version_from_doc(d) for d in docs]
+
+
+async def get_terms_page(
+    version_id: str,
+    search: str | None = None,
+    page: int = 1,
+    page_size: int = 50,
+) -> tuple[list[GlossaryTerm], int]:
+    """Returns (terms, total_count) for paginated UI preview."""
+    db = await get_database()
+    query: dict = {"version_id": version_id}
+    if search:
+        query["source_term_lower"] = {"$regex": re.escape(search.lower())}
+
+    total = await db[_COLL_TERMS].count_documents(query)
+    cursor = db[_COLL_TERMS].find(
+        query,
+        {"_id": 1, "source_term": 1, "translations": 1},
+        skip=(page - 1) * page_size,
+        limit=page_size,
+        sort=[("source_term_lower", 1)],
+    )
+    docs = await cursor.to_list(length=page_size)
+    terms = []
+    for d in docs:
+        d["_id"] = str(d["_id"])
+        terms.append(GlossaryTerm.model_validate(d))
+    return terms, total
diff --git a/backend/app/services/tts.py b/backend/app/services/tts.py
index dd66e35..861d9e6 100644
--- a/backend/app/services/tts.py
+++ b/backend/app/services/tts.py
@@ -74,19 +74,16 @@ class TTSService:
         # Determine which provider to use
         active_provider = provider or settings.tts_provider
 
-        # Extract simple language code for Gemini (e.g., "en-US" -> "en")
-        simple_lang = language_code.split("-")[0] if "-" in language_code else language_code
-
         # Try the configured provider first, then fallback
         if active_provider == "gemini" and self.gemini_available:
             try:
                 logger.info(
-                    f"Using Gemini TTS for language: {simple_lang}, voice: {voice_name}, "
+                    f"Using Gemini TTS for language: {language_code}, voice: {voice_name}, "
                     f"model: {model}, speed: {speed}x"
                 )
                 return await gemini_tts_service.synthesize_audio_description(
                     ad_vtt_content,
-                    simple_lang,
+                    language_code,
                     voice_name,
                     model=model,
                     speed=speed,
@@ -135,9 +132,6 @@ class TTSService:
         # Determine which provider to use
         active_provider = provider or settings.tts_provider
 
-        # Extract simple language code for Gemini (e.g., "en-US" -> "en")
-        simple_lang = language_code.split("-")[0] if "-" in language_code else language_code
-
         # Parse VTT cues first
         cues = self._parse_ad_cues(ad_vtt_content)
         if not cues:
@@ -169,7 +163,7 @@ class TTSService:
                     if active_provider == "gemini" and self.gemini_available:
                         audio_data = await gemini_tts_service.synthesize_text(
                             text, voice_name or gemini_tts_service.default_voice,
-                            simple_lang, model=model, speed=speed, style_prompt=style_prompt
+                            language_code, model=model, speed=speed, style_prompt=style_prompt
                         )
                     elif self.google_client:
                         audio_data = await self._synthesize_text_google(text, language_code, voice_name)
diff --git a/backend/app/tasks/__init__.py b/backend/app/tasks/__init__.py
index de9d225..14e0287 100644
--- a/backend/app/tasks/__init__.py
+++ b/backend/app/tasks/__init__.py
@@ -128,6 +128,7 @@ def import_task_modules():
         from . import notify  # noqa: E402, F401
         from . import ffmpeg_operations  # noqa: E402, F401
         from . import whisper_transcribe  # noqa: E402, F401
+        from . import embed_glossary  # noqa: E402, F401
         logger.info("Successfully imported all task modules")
     except Exception as e:
         logger.error(f"Error importing task modules: {e}")
diff --git a/backend/app/tasks/embed_glossary.py b/backend/app/tasks/embed_glossary.py
new file mode 100644
index 0000000..da90f20
--- /dev/null
+++ b/backend/app/tasks/embed_glossary.py
@@ -0,0 +1,102 @@
+"""
+Celery task: compute and store Gemini embeddings for all terms in a glossary version.
+
+Runs as a background job after glossary ingestion so the API response is fast.
+Processes terms in batches of 100 and updates embedded_count incrementally.
+"""
+from __future__ import annotations
+
+import asyncio
+
+from bson import ObjectId
+from motor.motor_asyncio import AsyncIOMotorClient
+
+from ..core.config import settings
+from ..core.logging import get_logger
+from ..models.glossary import EmbeddingStatus
+from . import celery_app
+
+logger = get_logger(__name__)
+
+_BATCH_SIZE = 100
+
+
+@celery_app.task(name="embed_glossary_version", bind=True, max_retries=3)
+def embed_glossary_version_task(self, version_id: str) -> dict:
+    """
+    Compute embeddings for all GlossaryTerms of `version_id`.
+    Updates embedded_count and embedding_status on the GlossaryVersion doc.
+    """
+    try:
+        result = asyncio.run(_async_embed_version(version_id))
+        return result
+    except Exception as exc:
+        logger.error(f"embed_glossary_version_task failed for {version_id}: {exc}")
+        raise self.retry(exc=exc, countdown=60) from None
+
+
+async def _async_embed_version(version_id: str) -> dict:
+    from ..services.embedding_service import embedding_service
+
+    mongo_client = AsyncIOMotorClient(settings.mongodb_uri)
+    db = mongo_client[settings.mongodb_db]
+
+    try:
+        # Mark in-progress
+        await db.glossary_versions.update_one(
+            {"_id": ObjectId(version_id)},
+            {"$set": {"embedding_status": EmbeddingStatus.IN_PROGRESS.value}},
+        )
+
+        # Fetch all terms without embeddings
+        cursor = db.glossary_terms.find(
+            {"version_id": version_id, "embedding": None},
+            {"_id": 1, "source_term": 1},
+        )
+        terms = await cursor.to_list(length=None)
+        total = len(terms)
+        logger.info(f"Embedding {total} terms for version {version_id}")
+
+        embedded_count = 0
+        for i in range(0, total, _BATCH_SIZE):
+            batch = terms[i: i + _BATCH_SIZE]
+            texts = [t["source_term"] for t in batch]
+            ids = [t["_id"] for t in batch]
+
+            embeddings = await embedding_service.embed_texts(texts)
+
+            # Bulk update
+            ops = []
+            from pymongo import UpdateOne
+            for term_id, embedding in zip(ids, embeddings, strict=False):
+                ops.append(UpdateOne({"_id": term_id}, {"$set": {"embedding": embedding}}))
+
+            if ops:
+                await db.glossary_terms.bulk_write(ops, ordered=False)
+
+            embedded_count += len(batch)
+            await db.glossary_versions.update_one(
+                {"_id": ObjectId(version_id)},
+                {"$set": {"embedded_count": embedded_count}},
+            )
+            logger.info(f"Version {version_id}: embedded {embedded_count}/{total}")
+
+        # Mark done
+        await db.glossary_versions.update_one(
+            {"_id": ObjectId(version_id)},
+            {"$set": {
+                "embedding_status": EmbeddingStatus.DONE.value,
+                "embedded_count": total,
+            }},
+        )
+        logger.info(f"Embedding complete for version {version_id}: {total} terms")
+        return {"version_id": version_id, "total": total}
+
+    except Exception:
+        await db.glossary_versions.update_one(
+            {"_id": ObjectId(version_id)},
+            {"$set": {"embedding_status": EmbeddingStatus.FAILED.value}},
+        )
+        raise
+    finally:
+        mongo_client.close()
diff --git a/backend/app/tasks/translate_and_synthesize.py b/backend/app/tasks/translate_and_synthesize.py
index 674f45a..4a84a32 100644
--- a/backend/app/tasks/translate_and_synthesize.py
+++ b/backend/app/tasks/translate_and_synthesize.py
@@ -219,6 +219,9 @@ async def _async_translate_and_synthesize(job_id: str):
 
         # Get translation mode (default to "traditional" for backwards compatibility)
         translation_mode = job_doc["requested_outputs"].get("translation_mode", "traditional")
+
+        # Glossary: lazy-loaded per target language during the loop
+        from ..services.glossary_service import get_glossary_block_for_job
         logger.info(f"Translation mode for job {job_id}: {translation_mode}")
 
         sdh_requested = job_doc["requested_outputs"].get("sdh_vtt", False)
@@ -293,12 +296,17 @@ async def _async_translate_and_synthesize(job_id: str):
                                 project_id=_cost_ctx["project_id"],
                             )
 
+                            # Build glossary block from source VTT for this language
+                            _job_for_glossary = {**job_doc, "_glossary_source_text": ""}
+                            _glossary = await get_glossary_block_for_job(_job_for_glossary, lang, db)
+
                             async def extract_targeted():
                                 return await gemini_service.extract_accessibility_targeted(
                                     video_local_path,
                                     lang,
                                     brand_context=job_brand_context,
                                     sdh_requested=sdh_requested,
+                                    glossary_block=_glossary,
                                     _cost_ctx=_cost_ctx,
                                 )
 
@@ -382,6 +390,9 @@ async def _async_translate_and_synthesize(job_id: str):
                                 logger.info(f"Successfully processed VTT files for language: {lang} (origin: video_native)")
 
             else:
+                # Combine source VTTs for glossary term matching
+                _source_text_for_glossary = " ".join(filter(None, [source_captions_vtt, source_ad_vtt]))
+
                 # TRADITIONAL MODE: Process languages sequentially
                 for language in target_languages:
                     logger.info(f"Processing language: {language} (from source: {source_language}, mode: {translation_mode})")
@@ -392,6 +403,10 @@ async def _async_translate_and_synthesize(job_id: str):
                         project_id=_cost_ctx["project_id"],
                     )
 
+                    # Lookup glossary terms for this target language
+                    _job_for_glossary = {**job_doc, "_glossary_source_text": _source_text_for_glossary}
+                    _glossary = await get_glossary_block_for_job(_job_for_glossary, language, db)
+
                     try:
                         if language in transcreation_languages:
                             # TRADITIONAL MODE with transcreation: cultural adaptation
@@ -401,6 +416,7 @@ async def _async_translate_and_synthesize(job_id: str):
                                     source_ad_vtt,
                                     language,
                                     brief="Standard accessibility content",
+                                    glossary_block=_glossary,
                                     _cost_ctx=_cost_ctx,
                                 )
 
@@ -414,12 +430,14 @@ async def _async_translate_and_synthesize(job_id: str):
                             async def translate_captions():
                                 return await gemini_service.translate_vtt(
                                     source_captions_vtt, language, source_language=source_language,
+                                    glossary_block=_glossary,
                                     _cost_ctx=_cost_ctx,
                                 )
 
                             async def translate_ad():
                                 return await gemini_service.translate_vtt(
                                     source_ad_vtt, language, source_language=source_language,
+                                    glossary_block=_glossary,
                                     _cost_ctx=_cost_ctx,
                                 )
 
@@ -448,6 +466,7 @@ async def _async_translate_and_synthesize(job_id: str):
                             async def translate_sdh():
                                 return await gemini_service.translate_vtt(
                                     source_sdh_vtt, language, source_language=source_language,
+                                    glossary_block=_glossary,
                                     _cost_ctx=_cost_ctx,
                                 )
                             translated_sdh = await retry_with_backoff(translate_sdh, max_retries=3)
diff --git a/backend/pyproject.toml b/backend/pyproject.toml
index 943a1b2..bda03bf 100644
--- a/backend/pyproject.toml
+++ b/backend/pyproject.toml
@@ -42,6 +42,8 @@ python-magic = "^0.4.27"
 aiohttp = "^3.12.15"
 jinja2 = "^3.1.6"
 audioop-lts = {version = "^0.2.2", python = ">=3.13"}
+openpyxl = "^3.1.2"
+pyahocorasick = "^2.1.1"
 
 [tool.poetry.group.dev.dependencies]
 pytest = "^7.4.3"
diff --git a/frontend/src/App.tsx b/frontend/src/App.tsx
index d49867e..d444ced 100644
--- a/frontend/src/App.tsx
+++ b/frontend/src/App.tsx
@@ -18,6 +18,9 @@ import { UserList } from './routes/admin/UserList';
 import { UserDetail } from './routes/admin/UserDetail';
 import { ClientList } from './routes/admin/ClientList';
 import { ClientDetail } from './routes/admin/ClientDetail';
+import { GlossaryList } from './routes/admin/glossaries/GlossaryList';
+import { GlossaryUpload } from './routes/admin/glossaries/GlossaryUpload';
+import { GlossaryDetail } from './routes/admin/glossaries/GlossaryDetail';
 import { AuditLog } from './routes/admin/AuditLog';
 import { LinguistQueue } from './routes/jobs/LinguistQueue';
 import { Downloads } from './routes/Downloads';
@@ -149,6 +152,27 @@ function AppContent() {
                 </RoleGate>
               </AuthenticatedRoute>
             } />
+            <Route path="/admin/clients/:clientId/glossaries" element={
+              <AuthenticatedRoute>
+                <RoleGate allowedRoles={['admin', 'project_manager', 'linguist', 'reviewer', 'production']}>
+                  <GlossaryList />
+                </RoleGate>
+              </AuthenticatedRoute>
+            } />
+            <Route path="/admin/clients/:clientId/glossaries/upload" element={
+              <AuthenticatedRoute>
+                <RoleGate allowedRoles={['admin', 'project_manager']}>
+                  <GlossaryUpload />
+                </RoleGate>
+              </AuthenticatedRoute>
+            } />
+            <Route path="/admin/clients/:clientId/glossaries/:glossaryId" element={
+              <AuthenticatedRoute>
+                <RoleGate allowedRoles={['admin', 'project_manager', 'linguist', 'reviewer', 'production']}>
+                  <GlossaryDetail />
+                </RoleGate>
+              </AuthenticatedRoute>
+            } />
             <Route path="/admin/audit-log" element={
               <AuthenticatedRoute>
                 <RoleGate allowedRoles={['production', 'admin']}>
diff --git a/frontend/src/lib/api.ts b/frontend/src/lib/api.ts
index b506dd7..7995106 100644
--- a/frontend/src/lib/api.ts
+++ b/frontend/src/lib/api.ts
@@ -59,6 +59,10 @@ import type {
   LanguageQCMapResponse,
   LanguageQCStateResponse,
   QueueResponse,
+  Glossary,
+  GlossaryDetail,
+  GlossaryVersion,
+  GlossaryTermsResponse,
 } from '../types/api';
 
 const API_BASE_URL = import.meta.env.VITE_API_BASE_URL || 'http://localhost:8000';
@@ -761,6 +765,84 @@ class ApiClient {
     const r = await this.client.get(`/me/language-qc-queue?${params.toString()}`);
     return r.data;
   }
+
+  // ── Glossary endpoints ──────────────────────────────────────────────────────
+
+  async getGlossaries(clientId: string): Promise<Glossary[]> {
+    const r = await this.client.get(`/clients/${clientId}/glossaries`);
+    return r.data;
+  }
+
+  async getGlossary(clientId: string, glossaryId: string): Promise<GlossaryDetail> {
+    const r = await this.client.get(`/clients/${clientId}/glossaries/${glossaryId}`);
+    return r.data;
+  }
+
+  async uploadGlossary(
+    clientId: string,
+    file: File,
+    name: string,
+    sourceLocale: string,
+    sourceLocaleCol: string,
+    description?: string,
+    changeNote?: string,
+  ): Promise<GlossaryDetail> {
+    const form = new FormData();
+    form.append('file', file);
+    form.append('name', name);
+    form.append('source_locale', sourceLocale);
+    form.append('source_locale_col', sourceLocaleCol);
+    if (description) form.append('description', description);
+    if (changeNote) form.append('change_note', changeNote);
+    const r = await this.client.post(`/clients/${clientId}/glossaries`, form, {
+      headers: { 'Content-Type': 'multipart/form-data' },
+      timeout: 120000,
+    });
+    return r.data;
+  }
+
+  async uploadGlossaryVersion(
+    clientId: string,
+    glossaryId: string,
+    file: File,
+    sourceLocaleCol: string,
+    changeNote?: string,
+  ): Promise<GlossaryVersion> {
+    const form = new FormData();
+    form.append('file', file);
+    form.append('source_locale_col', sourceLocaleCol);
+    if (changeNote) form.append('change_note', changeNote);
+    const r = await this.client.post(`/clients/${clientId}/glossaries/${glossaryId}/versions`, form, {
+      headers: { 'Content-Type': 'multipart/form-data' },
+      timeout: 120000,
+    });
+    return r.data;
+  }
+
+  async activateGlossaryVersion(clientId: string, glossaryId: string, versionId: string): Promise<{ status: string; active_version_id: string }> {
+    const form = new FormData();
+    form.append('version_id', versionId);
+    const r = await this.client.post(`/clients/${clientId}/glossaries/${glossaryId}/activate`, form);
+    return r.data;
+  }
+
+  async getGlossaryTerms(
+    clientId: string,
+    glossaryId: string,
+    opts?: { versionId?: string; search?: string; page?: number; pageSize?: number },
+  ): Promise<GlossaryTermsResponse> {
+    const params = new URLSearchParams();
+    if (opts?.versionId) params.append('version_id', opts.versionId);
+    if (opts?.search) params.append('search', opts.search);
+    if (opts?.page) params.append('page', String(opts.page));
+    if (opts?.pageSize) params.append('page_size', String(opts.pageSize));
+    const r = await this.client.get(`/clients/${clientId}/glossaries/${glossaryId}/terms?${params.toString()}`);
+    return r.data;
+  }
+
+  async archiveGlossary(clientId: string, glossaryId: string): Promise<void> {
+    await this.client.delete(`/clients/${clientId}/glossaries/${glossaryId}`);
+  }
 }
 
 export const apiClient = new ApiClient();
diff --git a/frontend/src/routes/admin/ClientDetail.tsx b/frontend/src/routes/admin/ClientDetail.tsx
index 84c2875..43ae232 100644
--- a/frontend/src/routes/admin/ClientDetail.tsx
+++ b/frontend/src/routes/admin/ClientDetail.tsx
@@ -1,5 +1,8 @@
 import { useState } from 'react';
-import { useParams } from 'react-router-dom';
+import { useParams, Link } from 'react-router-dom';
+import { useQuery } from '@tanstack/react-query';
+import { apiClient } from '../../lib/api';
+import type { Glossary } from '../../types/api';
 import {
   useClient,
   useTeams, useCreateTeam, useUpdateTeam, useDeleteTeam,
@@ -50,6 +53,12 @@ export function ClientDetail() {
 
   const [pmUserId, setPmUserId] = useState('');
 
+  const { data: glossaries = [] } = useQuery<Glossary[]>({
+    queryKey: ['glossaries', clientId],
+    queryFn: () => apiClient.getGlossaries(clientId!),
+    enabled: !!clientId,
+  });
+
   if (clientLoading) {
     return <div className="container mx-auto px-4 py-8 animate-pulse"><div className="h-8 bg-gray-200 rounded w-1/3" /></div>;
   }
@@ -337,6 +346,53 @@ export function ClientDetail() {
         </form>
       </section>
 
+      {/* Glossaries */}
+      <section className="bg-white rounded-xl border border-gray-200 p-5">
+        <div className="flex items-center justify-between mb-4">
+          <h2 className="text-base font-semibold text-gray-800">Glossaries</h2>
+          <Link
+            to={`/admin/clients/${clientId}/glossaries`}
+            className="text-sm text-blue-600 hover:text-blue-700"
+          >
+            View all →
+          </Link>
+        </div>
+        {glossaries.length === 0 ? (
+          <p className="text-sm text-gray-400">No glossaries yet</p>
+        ) : (
+          <div className="space-y-2">
+            {glossaries.slice(0, 3).map(g => (
+              <div key={g.id} className="flex items-center justify-between py-1.5">
+                <Link
+                  to={`/admin/clients/${clientId}/glossaries/${g.id}`}
+                  className="text-sm text-gray-800 hover:text-blue-600"
+                >
+                  {g.name}
+                </Link>
+                <span className={`text-xs px-2 py-0.5 rounded-full font-medium ${
+                  g.status === 'active' ? 'bg-green-100 text-green-700' : 'bg-gray-100 text-gray-500'
+                }`}>
+                  {g.status}
+                </span>
+              </div>
+            ))}
+            {glossaries.length > 3 && (
+              <p className="text-xs text-gray-400">+{glossaries.length - 3} more</p>
+            )}
+          </div>
+        )}
+        {(isAdmin || user?.role === 'project_manager') && (
+          <div className="mt-4">
+            <Link
+              to={`/admin/clients/${clientId}/glossaries/upload`}
+              className="text-sm text-blue-600 hover:underline"
+            >
+              + Upload glossary
+            </Link>
+          </div>
+        )}
+      </section>
+
       {/* Rename team modal */}
       {editingTeam && (
         <div className="fixed inset-0 bg-black/50 flex items-center justify-center z-50">
diff --git a/frontend/src/routes/admin/glossaries/GlossaryDetail.tsx b/frontend/src/routes/admin/glossaries/GlossaryDetail.tsx
new file mode 100644
index 0000000..622126c
--- /dev/null
+++ b/frontend/src/routes/admin/glossaries/GlossaryDetail.tsx
@@ -0,0 +1,335 @@
+import { useState, useRef } from 'react';
+import { useParams, Link } from 'react-router-dom';
+import { useQuery, useMutation, useQueryClient } from '@tanstack/react-query';
+import { apiClient } from '../../../lib/api';
+import { useToastContext } from '../../../contexts/ToastContext';
+import { useAuthStore } from '../../../lib/auth';
+import type { GlossaryVersion, GlossaryDetail as GlossaryDetailType } from '../../../types/api';
+
+type Tab = 'terms' | 'versions';
+
+function EmbeddingPill({ v }: { v: GlossaryVersion }) {
+  const pct = v.term_count > 0 ? Math.round((v.embedded_count / v.term_count) * 100) : 0;
+  switch (v.embedding_status) {
+    case 'done': return <span className="text-xs text-green-600 font-medium">Embedded ({v.embedded_count}/{v.term_count})</span>;
+    case 'in_progress': return (
+      <span className="text-xs text-blue-600 animate-pulse font-medium">
+        Embedding… {pct}% ({v.embedded_count}/{v.term_count})
+      </span>
+    );
+    case 'failed': return <span className="text-xs text-red-500 font-medium">Embedding failed</span>;
+    default: return <span className="text-xs text-gray-400">Pending embedding</span>;
+  }
+}
+
+export function GlossaryDetail() {
+  const { clientId, glossaryId } = useParams<{ clientId: string; glossaryId: string }>();
+  const { user } = useAuthStore();
+  const toast = useToastContext();
+  const qc = useQueryClient();
+  const isAdmin = user?.role === 'admin';
+  const isPM = user?.role === 'project_manager';
+
+  const [tab, setTab] = useState<Tab>('terms');
+  const [search, setSearch] = useState('');
+  const [page, setPage] = useState(1);
+
+  // New version upload state
+  const [showVersionUpload, setShowVersionUpload] = useState(false);
+  const [versionFile, setVersionFile] = useState<File | null>(null);
+  const [versionSourceCol, setVersionSourceCol] = useState('');
+  const [versionChangeNote, setVersionChangeNote] = useState('');
+  const versionFileRef = useRef<HTMLInputElement>(null);
+
+  const PAGE_SIZE = 50;
+
+  const { data: glossary, isLoading } = useQuery<GlossaryDetailType>({
+    queryKey: ['glossary', clientId, glossaryId],
+    queryFn: () => apiClient.getGlossary(clientId!, glossaryId!),
+    enabled: !!clientId && !!glossaryId,
+    refetchInterval: (q) => {
+      const g = q.state.data as GlossaryDetailType | undefined;
+      if (!g) return false;
+      const hasInProgress = g.versions.some(v => v.embedding_status === 'in_progress' || v.embedding_status === 'pending');
+      return hasInProgress ? 5000 : false;
+    },
+  });
+
+  const { data: termsData, isLoading: termsLoading } = useQuery({
+    queryKey: ['glossary-terms', clientId, glossaryId, search, page],
+    queryFn: () => apiClient.getGlossaryTerms(clientId!, glossaryId!, { search: search || undefined, page, pageSize: PAGE_SIZE }),
+    enabled: !!clientId && !!glossaryId && tab === 'terms',
+    placeholderData: (prev) => prev,
+  });
+
+  const activateMut = useMutation({
+    mutationFn: (versionId: string) => apiClient.activateGlossaryVersion(clientId!, glossaryId!, versionId),
+    onSuccess: () => {
+      qc.invalidateQueries({ queryKey: ['glossary', clientId, glossaryId] });
+      qc.invalidateQueries({ queryKey: ['glossaries', clientId] });
+      toast.success('Version activated');
+    },
+    onError: () => toast.error('Failed to activate version'),
+  });
+
+  const uploadVersionMut = useMutation({
+    mutationFn: () => apiClient.uploadGlossaryVersion(clientId!, glossaryId!, versionFile!, versionSourceCol.trim(), versionChangeNote.trim() || undefined),
+    onSuccess: () => {
+      qc.invalidateQueries({ queryKey: ['glossary', clientId, glossaryId] });
+      setShowVersionUpload(false);
+      setVersionFile(null);
+      setVersionSourceCol('');
+      setVersionChangeNote('');
+      toast.success('New version uploaded — embedding in background');
+    },
+    onError: (err: unknown) => {
+      const msg = (err as { response?: { data?: { detail?: string } } })?.response?.data?.detail ?? 'Upload failed';
+      toast.error(msg);
+    },
+  });
+
+  if (isLoading || !glossary) {
+    return (
+      <div className="container mx-auto px-4 py-8 max-w-4xl animate-pulse space-y-4">
+        <div className="h-8 bg-gray-200 rounded w-1/3" />
+        <div className="h-48 bg-gray-200 rounded-xl" />
+      </div>
+    );
+  }
+
+  const activeVersion = glossary.versions.find(v => v.id === glossary.current_version_id);
+  const totalPages = termsData ? Math.ceil(termsData.total / PAGE_SIZE) : 1;
+
+  return (
+    <div className="container mx-auto px-4 py-8 max-w-4xl space-y-6">
+      {/* Header */}
+      <div>
+        <p className="text-sm text-gray-400 mb-1">
+          ← <Link to={`/admin/clients/${clientId}/glossaries`} className="hover:text-blue-600">Glossaries</Link>
+        </p>
+        <div className="flex items-start justify-between gap-4">
+          <div>
+            <h1 className="text-2xl font-bold text-gray-900">{glossary.name}</h1>
+            {glossary.description && <p className="text-sm text-gray-500 mt-0.5">{glossary.description}</p>}
+            <p className="text-xs text-gray-400 mt-1">
+              Source: <span className="font-mono">{glossary.source_locale}</span>
+              {activeVersion && (
+                <> · Active: v{activeVersion.version_number} · {activeVersion.term_count.toLocaleString()} terms</>
+              )}
+            </p>
+            {activeVersion && (
+              <div className="mt-1"><EmbeddingPill v={activeVersion} /></div>
+            )}
+          </div>
+          {(isAdmin || isPM) && (
+            <button
+              onClick={() => setShowVersionUpload(!showVersionUpload)}
+              className="px-3 py-1.5 border border-gray-300 text-sm rounded-lg hover:bg-gray-50 shrink-0"
+            >
+              + New version
+            </button>
+          )}
+        </div>
+      </div>
+
+      {/* New version upload panel */}
+      {showVersionUpload && (
+        <div className="bg-blue-50 border border-blue-200 rounded-xl p-5 space-y-4">
+          <h3 className="text-sm font-semibold text-blue-900">Upload new version</h3>
+          <div
+            className="border-2 border-dashed border-blue-300 rounded-lg p-6 text-center cursor-pointer hover:border-blue-400"
+            onClick={() => versionFileRef.current?.click()}
+          >
+            <input
+              ref={versionFileRef}
+              type="file"
+              accept=".xlsx"
+              className="hidden"
+              onChange={(e) => setVersionFile(e.target.files?.[0] ?? null)}
+            />
+            {versionFile
+              ? <p className="text-sm font-medium text-gray-900">{versionFile.name}</p>
+              : <p className="text-sm text-gray-500">Click to select .xlsx file</p>
+            }
+          </div>
+          <div>
+            <label className="block text-xs font-medium text-gray-700 mb-1">Source column header *</label>
+            <input
+              type="text"
+              value={versionSourceCol}
+              onChange={e => setVersionSourceCol(e.target.value)}
+              placeholder="e.g. en_gb"
+              className="w-full border border-gray-300 rounded-lg px-3 py-1.5 text-sm focus:outline-none focus:ring-2 focus:ring-blue-500"
+            />
+          </div>
+          <div>
+            <label className="block text-xs font-medium text-gray-700 mb-1">Change note (optional)</label>
+            <input
+              type="text"
+              value={versionChangeNote}
+              onChange={e => setVersionChangeNote(e.target.value)}
+              placeholder="e.g. Updated Q2 terms"
+              className="w-full border border-gray-300 rounded-lg px-3 py-1.5 text-sm focus:outline-none focus:ring-2 focus:ring-blue-500"
+            />
+          </div>
+          <div className="flex gap-2">
+            <button
+              type="button"
+              onClick={() => setShowVersionUpload(false)}
+              className="px-3 py-1.5 border border-gray-300 text-sm rounded-lg hover:bg-gray-50"
+            >
+              Cancel
+            </button>
+            <button
+              type="button"
+              disabled={!versionFile || !versionSourceCol.trim() || uploadVersionMut.isPending}
+              onClick={() => uploadVersionMut.mutate()}
+              className="px-4 py-1.5 bg-blue-600 text-white text-sm font-medium rounded-lg hover:bg-blue-700 disabled:opacity-50"
+            >
+              {uploadVersionMut.isPending ? 'Uploading…' : 'Upload'}
+            </button>
+          </div>
+        </div>
+      )}
+
+      {/* Tabs */}
+      <div className="border-b border-gray-200">
+        <nav className="flex gap-6">
+          {(['terms', 'versions'] as Tab[]).map(t => (
+            <button
+              key={t}
+              onClick={() => setTab(t)}
+              className={`pb-3 text-sm font-medium border-b-2 transition-colors capitalize ${
+                tab === t ? 'border-blue-600 text-blue-600' : 'border-transparent text-gray-500 hover:text-gray-700'
+              }`}
+            >
+              {t}
+              {t === 'versions' && <span className="ml-1 text-xs text-gray-400">({glossary.versions.length})</span>}
+            </button>
+          ))}
+        </nav>
+      </div>
+
+      {/* Terms tab */}
+      {tab === 'terms' && (
+        <div className="space-y-4">
+          <input
+            type="text"
+            value={search}
+            onChange={e => { setSearch(e.target.value); setPage(1); }}
+            placeholder="Search terms…"
+            className="w-full border border-gray-300 rounded-lg px-3 py-2 text-sm focus:outline-none focus:ring-2 focus:ring-blue-500"
+          />
+
+          {termsLoading ? (
+            <div className="animate-pulse space-y-2">
+              {[1,2,3,4,5].map(i => <div key={i} className="h-10 bg-gray-100 rounded" />)}
+            </div>
+          ) : termsData?.terms.length === 0 ? (
+            <p className="text-sm text-gray-400 text-center py-8">
+              {search ? 'No terms match your search' : 'No terms in this glossary yet'}
+            </p>
+          ) : (
+            <>
+              <div className="text-xs text-gray-400">{termsData?.total.toLocaleString()} terms total</div>
+              <div className="border border-gray-200 rounded-xl overflow-hidden">
+                <table className="w-full text-sm">
+                  <thead className="bg-gray-50 border-b border-gray-200">
+                    <tr>
+                      <th className="text-left px-4 py-2.5 font-medium text-gray-600 w-1/3">Source term</th>
+                      <th className="text-left px-4 py-2.5 font-medium text-gray-600">Translations</th>
+                    </tr>
+                  </thead>
+                  <tbody className="divide-y divide-gray-100">
+                    {termsData?.terms.map((term, idx) => (
+                      <tr key={idx} className="hover:bg-gray-50">
+                        <td className="px-4 py-2.5 font-medium text-gray-900 align-top">{term.source_term}</td>
+                        <td className="px-4 py-2.5 text-gray-600">
+                          <div className="flex flex-wrap gap-2">
+                            {Object.entries(term.translations).slice(0, 6).map(([locale, text]) => (
+                              <span key={locale} className="inline-flex items-center gap-1 text-xs bg-gray-100 rounded px-1.5 py-0.5">
+                                <span className="font-mono text-gray-400">{locale}</span>
+                                <span className="text-gray-700 truncate max-w-[120px]">{text}</span>
+                              </span>
+                            ))}
+                            {Object.keys(term.translations).length > 6 && (
+                              <span className="text-xs text-gray-400">+{Object.keys(term.translations).length - 6} more</span>
+                            )}
+                          </div>
+                        </td>
+                      </tr>
+                    ))}
+                  </tbody>
+                </table>
+              </div>
+
+              {totalPages > 1 && (
+                <div className="flex items-center justify-between pt-2">
+                  <button
+                    disabled={page <= 1}
+                    onClick={() => setPage(p => p - 1)}
+                    className="px-3 py-1 text-sm border rounded disabled:opacity-40"
+                  >
+                    ← Previous
+                  </button>
+                  <span className="text-sm text-gray-500">Page {page} of {totalPages}</span>
+                  <button
+                    disabled={page >= totalPages}
+                    onClick={() => setPage(p => p + 1)}
+                    className="px-3 py-1 text-sm border rounded disabled:opacity-40"
+                  >
+                    Next →
+                  </button>
+                </div>
+              )}
+            </>
+          )}
+        </div>
+      )}
+
+      {/* Versions tab */}
+      {tab === 'versions' && (
+        <div className="space-y-3">
+          {glossary.versions.map((v) => {
+            const isActive = v.id === glossary.current_version_id;
+            return (
+              <div
+                key={v.id}
+                className={`rounded-xl border p-4 flex items-start justify-between gap-4 ${
+                  isActive ? 'border-blue-300 bg-blue-50' : 'border-gray-200 bg-white'
+                }`}
+              >
+                <div>
+                  <div className="flex items-center gap-2 mb-1">
+                    <span className="text-sm font-semibold text-gray-900">Version {v.version_number}</span>
+                    {isActive && (
+                      <span className="text-xs bg-blue-600 text-white px-2 py-0.5 rounded-full font-medium">Active</span>
+                    )}
+                  </div>
+                  <p className="text-xs text-gray-500">
+                    {v.term_count.toLocaleString()} terms · uploaded {new Date(v.created_at).toLocaleDateString()}
+                  </p>
+                  {v.change_note && <p className="text-xs text-gray-400 mt-0.5 italic">"{v.change_note}"</p>}
+                  <div className="mt-1"><EmbeddingPill v={v} /></div>
+                </div>
+                {(isAdmin || isPM) && !isActive && (
+                  <button
+                    onClick={() => {
+                      if (confirm(`Activate version ${v.version_number}? AI translations will start using this version.`)) {
+                        activateMut.mutate(v.id);
+                      }
+                    }}
+                    disabled={activateMut.isPending}
+                    className="text-xs px-3 py-1.5 bg-blue-600 text-white rounded-lg hover:bg-blue-700 disabled:opacity-50 shrink-0"
+                  >
+                    Activate
+                  </button>
+                )}
+              </div>
+            );
+          })}
+        </div>
+      )}
+    </div>
+  );
+}
diff --git a/frontend/src/routes/admin/glossaries/GlossaryList.tsx b/frontend/src/routes/admin/glossaries/GlossaryList.tsx
new file mode 100644
index 0000000..31c717a
--- /dev/null
+++ b/frontend/src/routes/admin/glossaries/GlossaryList.tsx
@@ -0,0 +1,131 @@
+import { useQuery, useMutation, useQueryClient } from '@tanstack/react-query';
+import { Link, useParams } from 'react-router-dom';
+import { apiClient } from '../../../lib/api';
+import { useToastContext } from '../../../contexts/ToastContext';
+import { useAuthStore } from '../../../lib/auth';
+import type { Glossary } from '../../../types/api';
+
+function statusBadge(status: string) {
+  return status === 'active'
+    ? 'bg-green-100 text-green-700'
+    : 'bg-gray-100 text-gray-500';
+}
+
+function embeddingBadge(status: string) {
+  switch (status) {
+    case 'done': return <span className="text-xs text-green-600">Embedded ✓</span>;
+    case 'in_progress': return <span className="text-xs text-blue-600 animate-pulse">Embedding…</span>;
+    case 'failed': return <span className="text-xs text-red-500">Embed failed</span>;
+    default: return <span className="text-xs text-gray-400">Pending embed</span>;
+  }
+}
+
+export function GlossaryList() {
+  const { clientId } = useParams<{ clientId: string }>();
+  const { user } = useAuthStore();
+  const toast = useToastContext();
+  const qc = useQueryClient();
+  const isAdmin = user?.role === 'admin';
+  const isPM = user?.role === 'project_manager';
+
+  const { data: glossaries = [], isLoading } = useQuery<Glossary[]>({
+    queryKey: ['glossaries', clientId],
+    queryFn: () => apiClient.getGlossaries(clientId!),
+    enabled: !!clientId,
+    refetchInterval: (q) => {
+      const data = q.state.data;
+      return Array.isArray(data) ? 5000 : false;
+    },
+  });
+
+  const archiveMut = useMutation({
+    mutationFn: (id: string) => apiClient.archiveGlossary(clientId!, id),
+    onSuccess: () => {
+      qc.invalidateQueries({ queryKey: ['glossaries', clientId] });
+      toast.success('Glossary archived');
+    },
+    onError: () => toast.error('Failed to archive glossary'),
+  });
+
+  if (isLoading) {
+    return (
+      <div className="container mx-auto px-4 py-8 max-w-4xl animate-pulse space-y-3">
+        {[1, 2].map(i => <div key={i} className="h-16 bg-gray-200 rounded-xl" />)}
+      </div>
+    );
+  }
+
+  return (
+    <div className="container mx-auto px-4 py-8 max-w-4xl space-y-6">
+      <div className="flex items-center justify-between">
+        <div>
+          <p className="text-sm text-gray-400 mb-1">
+            ← <Link to={`/admin/clients/${clientId}`} className="hover:text-blue-600">Client</Link>
+          </p>
+          <h1 className="text-2xl font-bold text-gray-900">Glossaries</h1>
+        </div>
+        {(isAdmin || isPM) && (
+          <Link
+            to={`/admin/clients/${clientId}/glossaries/upload`}
+            className="px-4 py-2 bg-blue-600 text-white text-sm font-medium rounded-lg hover:bg-blue-700"
+          >
+            + Upload glossary
+          </Link>
+        )}
+      </div>
+
+      {glossaries.length === 0 ? (
+        <div className="text-center py-16 text-gray-400">
+          <p className="text-lg mb-2">No glossaries yet</p>
+          {(isAdmin || isPM) && (
+            <Link to={`/admin/clients/${clientId}/glossaries/upload`} className="text-blue-500 hover:underline text-sm">
+              Upload the first glossary
+            </Link>
+          )}
+        </div>
+      ) : (
+        <div className="space-y-3">
+          {glossaries.map((g) => (
+            <div key={g.id} className="bg-white rounded-xl border border-gray-200 p-5 flex items-start justify-between gap-4">
+              <div className="flex-1 min-w-0">
+                <div className="flex items-center gap-2 mb-1">
+                  <Link
+                    to={`/admin/clients/${clientId}/glossaries/${g.id}`}
+                    className="text-base font-semibold text-gray-900 hover:text-blue-600 truncate"
+                  >
+                    {g.name}
+                  </Link>
+                  <span className={`text-xs px-2 py-0.5 rounded-full font-medium ${statusBadge(g.status)}`}>
+                    {g.status}
+                  </span>
+                </div>
+                {g.description && <p className="text-sm text-gray-500 mb-1 truncate">{g.description}</p>}
+                <p className="text-xs text-gray-400">
+                  Source: <span className="font-mono">{g.source_locale}</span>
+                  {' · '}Created {new Date(g.created_at).toLocaleDateString()}
+                </p>
+              </div>
+              <div className="flex items-center gap-4 shrink-0">
+                <div className="text-right text-xs text-gray-400">
+                  {g.current_version_id ? embeddingBadge('') : null}
+                </div>
+                {isAdmin && g.status === 'active' && (
+                  <button
+                    onClick={() => {
+                      if (confirm('Archive this glossary? It will no longer be used for AI translations.')) {
+                        archiveMut.mutate(g.id);
+                      }
+                    }}
+                    className="text-xs text-red-500 hover:text-red-700"
+                  >
+                    Archive
+                  </button>
+                )}
+              </div>
+            </div>
+          ))}
+        </div>
+      )}
+    </div>
+  );
+}
diff --git a/frontend/src/routes/admin/glossaries/GlossaryUpload.tsx b/frontend/src/routes/admin/glossaries/GlossaryUpload.tsx
new file mode 100644
index 0000000..362a6f3
--- /dev/null
+++ b/frontend/src/routes/admin/glossaries/GlossaryUpload.tsx
@@ -0,0 +1,204 @@
+import { useState, useRef } from 'react';
+import { useNavigate, useParams, Link } from 'react-router-dom';
+import { useMutation, useQueryClient } from '@tanstack/react-query';
+import { apiClient } from '../../../lib/api';
+import { useToastContext } from '../../../contexts/ToastContext';
+
+const KNOWN_LOCALES: { code: string; label: string }[] = [
+  { code: 'en-GB', label: 'English (UK)' },
+  { code: 'en-US', label: 'English (US)' },
+  { code: 'en-CA', label: 'English (Canada)' },
+  { code: 'de-DE', label: 'German' },
+  { code: 'fr-FR', label: 'French (France)' },
+  { code: 'fr-CA', label: 'French (Canada)' },
+  { code: 'es-ES', label: 'Spanish (Spain)' },
+  { code: 'es-MX', label: 'Spanish (Mexico)' },
+  { code: 'es-419', label: 'Spanish (Latin America)' },
+  { code: 'it-IT', label: 'Italian' },
+  { code: 'pt-BR', label: 'Portuguese (Brazil)' },
+  { code: 'pt-PT', label: 'Portuguese (Portugal)' },
+  { code: 'nl-NL', label: 'Dutch' },
+  { code: 'pl-PL', label: 'Polish' },
+  { code: 'cs-CZ', label: 'Czech' },
+  { code: 'tr-TR', label: 'Turkish' },
+  { code: 'ko-KR', label: 'Korean' },
+  { code: 'ja-JP', label: 'Japanese' },
+  { code: 'id-ID', label: 'Indonesian' },
+];
+
+export function GlossaryUpload() {
+  const { clientId } = useParams<{ clientId: string }>();
+  const navigate = useNavigate();
+  const toast = useToastContext();
+  const qc = useQueryClient();
+
+  const [file, setFile] = useState<File | null>(null);
+  const [name, setName] = useState('');
+  const [sourceLocale, setSourceLocale] = useState('en-GB');
+  const [sourceLocaleCol, setSourceLocaleCol] = useState('');
+  const [description, setDescription] = useState('');
+  const [changeNote, setChangeNote] = useState('');
+  const [dragOver, setDragOver] = useState(false);
+  const fileInputRef = useRef<HTMLInputElement>(null);
+
+  const uploadMut = useMutation({
+    mutationFn: () => apiClient.uploadGlossary(
+      clientId!,
+      file!,
+      name.trim(),
+      sourceLocale,
+      sourceLocaleCol.trim(),
+      description.trim() || undefined,
+      changeNote.trim() || undefined,
+    ),
+    onSuccess: (g) => {
+      qc.invalidateQueries({ queryKey: ['glossaries', clientId] });
+      toast.success(`Glossary "${g.name}" uploaded — embedding in background`);
+      navigate(`/admin/clients/${clientId}/glossaries/${g.id}`);
+    },
+    onError: (err: unknown) => {
+      const msg = (err as { response?: { data?: { detail?: string } } })?.response?.data?.detail ?? 'Upload failed';
+      toast.error(msg);
+    },
+  });
+
+  const handleDrop = (e: React.DragEvent) => {
+    e.preventDefault();
+    setDragOver(false);
+    const f = e.dataTransfer.files[0];
+    if (f && f.name.endsWith('.xlsx')) setFile(f);
+    else toast.error('Only .xlsx files are accepted');
+  };
+
+  const canSubmit = !!file && !!name.trim() && !!sourceLocale && !!sourceLocaleCol.trim() && !uploadMut.isPending;
+
+  return (
+    <div className="container mx-auto px-4 py-8 max-w-xl space-y-6">
+      <div>
+        <p className="text-sm text-gray-400 mb-1">
+          ← <Link to={`/admin/clients/${clientId}/glossaries`} className="hover:text-blue-600">Glossaries</Link>
+        </p>
+        <h1 className="text-2xl font-bold text-gray-900">Upload glossary</h1>
+        <p className="text-sm text-gray-500 mt-1">Upload an xlsx file with terminology translations.</p>
+      </div>
+
+      {/* Drop zone */}
+      <div
+        className={`border-2 border-dashed rounded-xl p-8 text-center cursor-pointer transition-colors ${
+          dragOver ? 'border-blue-400 bg-blue-50' : 'border-gray-300 hover:border-gray-400'
+        }`}
+        onClick={() => fileInputRef.current?.click()}
+        onDragOver={(e) => { e.preventDefault(); setDragOver(true); }}
+        onDragLeave={() => setDragOver(false)}
+        onDrop={handleDrop}
+      >
+        <input
+          ref={fileInputRef}
+          type="file"
+          accept=".xlsx,application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
+          className="hidden"
+          onChange={(e) => {
+            const f = e.target.files?.[0];
+            if (f) setFile(f);
+          }}
+        />
+        {file ? (
+          <div>
+            <p className="text-sm font-medium text-gray-900">{file.name}</p>
+            <p className="text-xs text-gray-400 mt-1">{(file.size / 1024 / 1024).toFixed(1)} MB</p>
+            <button
+              type="button"
+              onClick={(e) => { e.stopPropagation(); setFile(null); }}
+              className="mt-2 text-xs text-red-500 hover:text-red-700"
+            >
+              Remove
+            </button>
+          </div>
+        ) : (
+          <div className="text-gray-400">
+            <p className="text-sm font-medium">Drop .xlsx file here or click to browse</p>
+            <p className="text-xs mt-1">Max 50 MB</p>
+          </div>
+        )}
+      </div>
+
+      {/* Form fields */}
+      <div className="space-y-4">
+        <div>
+          <label className="block text-sm font-medium text-gray-700 mb-1">Glossary name *</label>
+          <input
+            type="text"
+            value={name}
+            onChange={e => setName(e.target.value)}
+            placeholder="e.g. 3M Master Terminology"
+            className="w-full border border-gray-300 rounded-lg px-3 py-2 text-sm focus:outline-none focus:ring-2 focus:ring-blue-500"
+          />
+        </div>
+
+        <div>
+          <label className="block text-sm font-medium text-gray-700 mb-1">Source language *</label>
+          <select
+            value={sourceLocale}
+            onChange={e => setSourceLocale(e.target.value)}
+            className="w-full border border-gray-300 rounded-lg px-3 py-2 text-sm focus:outline-none focus:ring-2 focus:ring-blue-500"
+          >
+            {KNOWN_LOCALES.map(l => (
+              <option key={l.code} value={l.code}>{l.label} ({l.code})</option>
+            ))}
+          </select>
+        </div>
+
+        <div>
+          <label className="block text-sm font-medium text-gray-700 mb-1">Source column header in xlsx *</label>
+          <input
+            type="text"
+            value={sourceLocaleCol}
+            onChange={e => setSourceLocaleCol(e.target.value)}
+            placeholder="e.g. en_gb  or  English (GB)"
+            className="w-full border border-gray-300 rounded-lg px-3 py-2 text-sm focus:outline-none focus:ring-2 focus:ring-blue-500"
+          />
+          <p className="text-xs text-gray-400 mt-1">Must exactly match the column header in the xlsx file (case-insensitive).</p>
+        </div>
+
+        <div>
+          <label className="block text-sm font-medium text-gray-700 mb-1">Description (optional)</label>
+          <textarea
+            value={description}
+            onChange={e => setDescription(e.target.value)}
+            rows={2}
+            className="w-full border border-gray-300 rounded-lg px-3 py-2 text-sm focus:outline-none focus:ring-2 focus:ring-blue-500 resize-none"
+          />
+        </div>
+
+        <div>
+          <label className="block text-sm font-medium text-gray-700 mb-1">Change note (optional)</label>
+          <input
+            type="text"
+            value={changeNote}
+            onChange={e => setChangeNote(e.target.value)}
+            placeholder="e.g. Initial import Q1 2026"
+            className="w-full border border-gray-300 rounded-lg px-3 py-2 text-sm focus:outline-none focus:ring-2 focus:ring-blue-500"
+          />
+        </div>
+      </div>
+
+      <div className="flex gap-3 pt-2">
+        <button
+          type="button"
+          onClick={() => navigate(`/admin/clients/${clientId}/glossaries`)}
+          className="px-4 py-2 border border-gray-300 text-sm rounded-lg hover:bg-gray-50"
+        >
+          Cancel
+        </button>
+        <button
+          type="button"
+          disabled={!canSubmit}
+          onClick={() => uploadMut.mutate()}
+          className="flex-1 px-4 py-2 bg-blue-600 text-white text-sm font-medium rounded-lg hover:bg-blue-700 disabled:opacity-50"
+        >
+          {uploadMut.isPending ? 'Uploading…' : 'Upload glossary'}
+        </button>
+      </div>
+    </div>
+  );
+}
diff --git a/frontend/src/types/api.ts b/frontend/src/types/api.ts
index 88298e0..f6c19ff 100644
--- a/frontend/src/types/api.ts
+++ b/frontend/src/types/api.ts
@@ -741,4 +741,60 @@ export interface AuditLogQuery {
   limit?: number;
   sort_by?: string;
   sort_order?: number;
+}
+
+// ── Glossary ────────────────────────────────────────────────────────────────
+
+export type GlossaryStatus = 'active' | 'archived';
+export type EmbeddingStatus = 'pending' | 'in_progress' | 'done' | 'failed';
+
+export interface GlossaryVersion {
+  id: string;
+  glossary_id: string;
+  version_number: number;
+  source_xlsx_gcs_path?: string;
+  term_count: number;
+  embedded_count: number;
+  embedding_status: EmbeddingStatus;
+  created_at: string;
+  created_by: string;
+  change_note?: string;
+}
+
+export interface Glossary {
+  id: string;
+  client_id: string;
+  name: string;
+  description?: string;
+  source_locale: string;
+  source: string;
+  status: GlossaryStatus;
+  current_version_id?: string;
+  created_at: string;
+  created_by: string;
+}
+
+export interface GlossaryDetail extends Glossary {
+  versions: GlossaryVersion[];
+}
+
+export interface GlossaryTerm {
+  id?: string;
+  source_term: string;
+  translations: Record<string, string>;
+}
+
+export interface GlossaryTermsResponse {
+  terms: GlossaryTerm[];
+  total: number;
+  page: number;
+  page_size: number;
+}
+
+export interface GlossaryUploadRequest {
+  name: string;
+  source_locale: string;
+  source_locale_col: string;
+  description?: string;
+  change_note?: string;
 }
\ No newline at end of file