diff --git a/server/app/api/tasks.py b/server/app/api/tasks.py
index 652b858..680e74c 100644
--- a/server/app/api/tasks.py
+++ b/server/app/api/tasks.py
@@ -10,7 +10,7 @@
 
 from ..auth import current_user
 from ..core.engine.engine import REGISTRY, latest_checkpoint, step_durations
-from ..core.engine.task_templates import TEMPLATES, apply_template
+from ..core.engine.task_templates import apply_template, localized_templates
 from ..db import get_db
 from ..models import Task, User
 
@@ -29,8 +29,9 @@ class TaskIn(BaseModel):
 
 
 @router.get("/templates")
-def list_templates():
-    return TEMPLATES
+def list_templates(lang: str = "en"):
+    """Task templates for the picker. ``lang`` (en/zh) localizes the display strings."""
+    return localized_templates("zh" if lang == "zh" else "en")
 
 
 def _parse_prompt(prompt: str) -> tuple[str, dict, str]:
diff --git a/server/app/core/engine/task_templates.py b/server/app/core/engine/task_templates.py
index 3b1f55b..e68dd1e 100644
--- a/server/app/core/engine/task_templates.py
+++ b/server/app/core/engine/task_templates.py
@@ -2,69 +2,130 @@
 
 The template wording targets the general pattern of "knowledge-work monitoring"; the current connectors are arXiv/Semantic Scholar
 (covering disciplines such as CS/physics/math/finance/biology/engineering), and other information sources extend along the same graph structure.
+
+Display strings (name/description/field label/placeholder) are bilingual: the API localizes them per request language via
+``localized_templates(lang)``. Functional keys (id/task_type/field key/type/default) stay language-independent.
 """
 from __future__ import annotations
 
+
+def _L(en: str, zh: str) -> dict:
+    """A bilingual display string. Resolved to one language by ``_pick`` at request time."""
+    return {"en": en, "zh": zh}
+
+
 TEMPLATES = [
     {
         "id": "literature_watch",
-        "name": "Topic watch",
-        "description": "Continuously monitor new publications on any topic — AI, finance, "
-                       "biology, engineering — auto-triage, deep-summarize, archive, brief",
+        "name": _L("Topic watch", "主题追踪"),
+        "description": _L(
+            "Continuously monitor new publications on any topic — AI, finance, "
+            "biology, engineering — auto-triage, deep-summarize, archive, brief",
+            "持续追踪任意主题的新成果——AI、金融、生物、工程——自动初筛、精读摘要、归档、生成简报",
+        ),
         "task_type": "arxiv_watch",
         "fields": [
-            {"key": "query", "label": "Topic keywords", "type": "text",
-             "placeholder": "e.g. LLM agent security / portfolio optimization"},
-            {"key": "max_results", "label": "Papers per poll", "type": "number", "default": 15},
+            {"key": "query", "type": "text",
+             "label": _L("Topic keywords", "主题关键词"),
+             "placeholder": _L("e.g. LLM agent security / portfolio optimization",
+                               "如 LLM agent security / 投资组合优化")},
+            {"key": "max_results", "type": "number", "default": 15,
+             "label": _L("Papers per poll", "每轮抓取篇数")},
         ],
     },
     {
         "id": "security_watch",
-        "name": "Security watch",
-        "description": "Track security research the way an on-call team tracks advisories — "
-                       "new attack & defense work on your stack, triaged and summarized",
+        "name": _L("Security watch", "安全追踪"),
+        "description": _L(
+            "Track security research the way an on-call team tracks advisories — "
+            "new attack & defense work on your stack, triaged and summarized",
+            "像值班团队盯安全通告一样追踪安全研究——你技术栈上的新攻防工作,自动初筛并摘要",
+        ),
         "task_type": "arxiv_watch",
         "fields": [
-            {"key": "query", "label": "What to guard", "type": "text",
-             "placeholder": "e.g. prompt injection defense / container escape"},
-            {"key": "max_results", "label": "Papers per poll", "type": "number", "default": 15},
+            {"key": "query", "type": "text",
+             "label": _L("What to guard", "关注的防护面"),
+             "placeholder": _L("e.g. prompt injection defense / container escape",
+                               "如 提示注入防御 / 容器逃逸")},
+            {"key": "max_results", "type": "number", "default": 15,
+             "label": _L("Papers per poll", "每轮抓取篇数")},
         ],
     },
     {
         "id": "market_watch",
-        "name": "Market & finance watch",
-        "description": "Track markets both ways — breaking financial news (stocks, earnings, "
-                       "M&A) and quantitative finance preprints — triaged and briefed every morning",
+        "name": _L("Market & finance watch", "市场与金融追踪"),
+        "description": _L(
+            "Track markets both ways — breaking financial news (stocks, earnings, "
+            "M&A) and quantitative finance preprints — triaged and briefed every morning",
+            "双向追踪市场——突发财经新闻(股票、财报、并购)与量化金融预印本——每早初筛并生成简报",
+        ),
         "task_type": "arxiv_watch",
         "fields": [
-            {"key": "query", "label": "Market topic or ticker", "type": "text",
-             "placeholder": "e.g. Micron MU stock / portfolio optimization"},
-            {"key": "max_results", "label": "Items per poll", "type": "number", "default": 15},
+            {"key": "query", "type": "text",
+             "label": _L("Market topic or ticker", "市场主题或股票代码"),
+             "placeholder": _L("e.g. Micron MU stock / portfolio optimization",
+                               "如 美光 MU 股票 / 投资组合优化")},
+            {"key": "max_results", "type": "number", "default": 15,
+             "label": _L("Items per poll", "每轮抓取条数")},
         ],
     },
     {
         "id": "proposal_research",
-        "name": "Deep-dive survey",
-        "description": "One focused sweep over a specific question — search wide, triage, "
-                       "summarize everything relevant. Great before starting a project",
+        "name": _L("Deep-dive survey", "深度调研"),
+        "description": _L(
+            "One focused sweep over a specific question — search wide, triage, "
+            "summarize everything relevant. Great before starting a project",
+            "针对一个具体问题做一次集中扫描——广泛检索、初筛、把相关内容全部摘要。开题前很好用",
+        ),
         "task_type": "arxiv_watch",
         "fields": [
-            {"key": "query", "label": "Research question", "type": "text",
-             "placeholder": "e.g. catastrophic forgetting in multi-agent RL"},
-            {"key": "max_results", "label": "Papers to sweep (go big)", "type": "number", "default": 30},
+            {"key": "query", "type": "text",
+             "label": _L("Research question", "研究问题"),
+             "placeholder": _L("e.g. catastrophic forgetting in multi-agent RL",
+                               "如 多智能体强化学习中的灾难性遗忘")},
+            {"key": "max_results", "type": "number", "default": 30,
+             "label": _L("Papers to sweep (go big)", "扫描篇数(可以大胆调大)")},
         ],
     },
     {
         "id": "weekly_briefing",
-        "name": "Briefing now",
-        "description": "Don't wait for tomorrow morning — aggregate the last 24h of "
-                       "summaries into a briefing right now",
+        "name": _L("Briefing now", "立即生成简报"),
+        "description": _L(
+            "Don't wait for tomorrow morning — aggregate the last 24h of "
+            "summaries into a briefing right now",
+            "不必等到明早——现在就把过去 24 小时的摘要汇总成一份简报",
+        ),
         "task_type": "briefing",
         "fields": [],
     },
 ]
 
 
+def _pick(value, lang: str):
+    """Resolve a bilingual dict to one language; pass through plain values unchanged."""
+    if isinstance(value, dict) and ("en" in value or "zh" in value):
+        return value.get(lang) or value.get("en") or ""
+    return value
+
+
+def localized_templates(lang: str = "en") -> list[dict]:
+    """Return templates with display strings resolved to ``lang`` (functional keys untouched)."""
+    out = []
+    for t in TEMPLATES:
+        out.append({
+            "id": t["id"],
+            "name": _pick(t["name"], lang),
+            "description": _pick(t["description"], lang),
+            "task_type": t["task_type"],
+            "fields": [
+                {k: (_pick(v, lang) if k in ("label", "placeholder") else v)
+                 for k, v in f.items()}
+                for f in t["fields"]
+            ],
+        })
+    return out
+
+
 def apply_template(template_id: str, values: dict) -> tuple[str, dict, str]:
     tpl = next((t for t in TEMPLATES if t["id"] == template_id), None)
     if tpl is None:
@@ -74,7 +135,7 @@ def apply_template(template_id: str, values: dict) -> tuple[str, dict, str]:
         val = values.get(f["key"], f.get("default", ""))
         if f["type"] == "text" and not str(val).strip():
             # text fields without a default are required: an empty query makes the arXiv API error out directly and the task is bound to fail
-            raise ValueError(f"Field '{f['label']}' is required")
+            raise ValueError(f"Field '{_pick(f['label'], 'en')}' is required")
         params[f["key"]] = val
-    title = f"{tpl['name']}: {params.get('query', '')}".strip(": ")[:60]
+    title = f"{_pick(tpl['name'], 'en')}: {params.get('query', '')}".strip(": ")[:60]
     return tpl["task_type"], params, title
diff --git a/web/app/deploy/page.tsx b/web/app/deploy/page.tsx
index 4d31653..3fd51c1 100644
--- a/web/app/deploy/page.tsx
+++ b/web/app/deploy/page.tsx
@@ -23,12 +23,38 @@ export default function Deploy() {
   const { lang } = useLang();
   const L = <T,>(en: T, zh: T): T => (lang === "zh" ? zh : en);
   const [copied, setCopied] = useState(false);
+  const [failed, setFailed] = useState(false);
 
   function copy() {
-    navigator.clipboard.writeText(PROMPT).then(() => {
+    const done = () => {
+      setFailed(false);
       setCopied(true);
       setTimeout(() => setCopied(false), 2500);
-    });
+    };
+    // navigator.clipboard only exists in secure contexts (HTTPS / localhost).
+    // The public demo is served over plain HTTP, where it is undefined — fall back to execCommand.
+    if (navigator.clipboard && window.isSecureContext) {
+      navigator.clipboard.writeText(PROMPT).then(done).catch(legacyCopy);
+    } else {
+      legacyCopy();
+    }
+
+    function legacyCopy() {
+      try {
+        const ta = document.createElement("textarea");
+        ta.value = PROMPT;
+        ta.setAttribute("readonly", "");
+        ta.style.position = "fixed";
+        ta.style.left = "-9999px";
+        document.body.appendChild(ta);
+        ta.select();
+        const ok = document.execCommand("copy");
+        document.body.removeChild(ta);
+        ok ? done() : setFailed(true);
+      } catch {
+        setFailed(true);
+      }
+    }
   }
 
   return (
@@ -50,6 +76,12 @@ export default function Deploy() {
             {copied ? L("✅ Copied!", "✅ 已复制!") : L("📋 Copy the prompt", "📋 一键复制")}
           </button>
         </div>
+        {failed && (
+          <p className="text-xs text-amber-600">
+            {L("Auto-copy is blocked here — select the text below and copy manually (Ctrl/Cmd+C).",
+               "此环境禁用了自动复制,请手动选中下方文本复制(Ctrl/Cmd+C)。")}
+          </p>
+        )}
         <pre className="max-h-96 overflow-auto whitespace-pre-wrap rounded-lg bg-slate-900 p-4 font-mono text-xs leading-5 text-slate-200">
 {PROMPT}
         </pre>
diff --git a/web/app/tasks/page.tsx b/web/app/tasks/page.tsx
index 144cbf4..450e331 100644
--- a/web/app/tasks/page.tsx
+++ b/web/app/tasks/page.tsx
@@ -36,9 +36,13 @@ export default function Tasks() {
 
   const load = () => api("/api/tasks").then(setTasks).catch(() => {});
 
+  // Templates carry backend-provided display copy; re-fetch on language change so the picker follows the UI language.
+  useEffect(() => {
+    api(`/api/tasks/templates?lang=${lang}`).then(setTemplates).catch(() => {});
+  }, [lang]);
+
   useEffect(() => {
     load();
-    api("/api/tasks/templates").then(setTemplates).catch(() => {});
     return subscribe((type, ev) => {
       if (type === "task_progress") {
         setTasks((ts) => (ts || []).map((t) => t.id === ev.task_id