diff --git a/CHANGELOG.md b/CHANGELOG.md
index 0cb0ac0..7e38c0d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,13 @@
 # Hermes Web UI -- Changelog
 
+## [v0.50.57] — 2026-04-15
+
+### Added
+- **Auto-generated session titles** — after the first exchange, a background thread generates a concise title from the first user message and assistant reply, replacing the default first-message substring. Updates live in the UI via a new `title` SSE event. Manual renames are preserved; generation only runs once per session. Includes MiniMax token budget handling and a local heuristic fallback. (Fixes #495, PR #535 by @franksong2702)
+
+### Changed
+- **SSE stream termination** — streams now end with `stream_end` instead of `done` so the background title generation thread has time to emit the title update before the client disconnects.
+
 ## [v0.50.55] — 2026-04-15
 
 ### Fixed
diff --git a/api/routes.py b/api/routes.py
index a27ae89..6361cd4 100644
--- a/api/routes.py
+++ b/api/routes.py
@@ -1459,7 +1459,7 @@ def _handle_sse_stream(handler, parsed):
                 handler.wfile.flush()
                 continue
             _sse(handler, event, data)
-            if event in ("done", "error", "cancel"):
+            if event in ("stream_end", "error", "cancel"):
                 break
     except (BrokenPipeError, ConnectionResetError):
         pass
diff --git a/api/streaming.py b/api/streaming.py
index f77ce82..213da1b 100644
--- a/api/streaming.py
+++ b/api/streaming.py
@@ -6,10 +6,12 @@ import json
 import logging
 import os
 import queue
+import re
 import threading
 import time
 import traceback
 from pathlib import Path
+from typing import Optional
 
 logger = logging.getLogger(__name__)
 
@@ -58,6 +60,442 @@ from api.workspace import set_last_workspace
 _API_SAFE_MSG_KEYS = {'role', 'content', 'tool_calls', 'tool_call_id', 'name', 'refusal'}
 
 
+def _strip_thinking_markup(text: str) -> str:
+    """Remove common reasoning/thinking wrappers from model text."""
+    if not text:
+        return ''
+    s = str(text)
+    s = re.sub(r'<think>.*?</think>', ' ', s, flags=re.IGNORECASE | re.DOTALL)
+    s = re.sub(r'<\|channel\|>thought.*?<channel\|>', ' ', s, flags=re.IGNORECASE | re.DOTALL)
+    s = re.sub(r'^\s*(the|ther)\s+user\s+is\s+asking.*$', ' ', s, flags=re.IGNORECASE | re.MULTILINE)
+    s = re.sub(r'\s+', ' ', s).strip()
+    return s
+
+
+def _sanitize_generated_title(text: str) -> str:
+    """Sanitize LLM-generated title text before persisting to session."""
+    s = _strip_thinking_markup(text or '')
+    s = re.sub(r'^\s*title\s*:\s*', '', s, flags=re.IGNORECASE)
+    s = s.strip(" \t\r\n\"'`")
+    s = re.sub(r'\s+', ' ', s).strip()
+    # Guard against chain-of-thought leakage and meta-reasoning patterns.
+    if _looks_invalid_generated_title(s):
+        return ''
+    return s[:80]
+
+
+def _looks_invalid_generated_title(text: str) -> bool:
+    s = str(text or '')
+    if not s.strip():
+        return True
+    return bool(
+        re.search(r'<think>|<\|channel\|>thought', s, flags=re.IGNORECASE)
+        or re.search(r'^\s*(the|ther)\s+user\s+', s, flags=re.IGNORECASE)
+        or re.search(r'^\s*user\s+\w+\s+', s, flags=re.IGNORECASE)
+        or re.search(r'\b(they|user)\s+want(s)?\s+me\s+to\b', s, flags=re.IGNORECASE)
+        or re.search(r'^\s*(i|we)\s+(should|need to|will|can)\b', s, flags=re.IGNORECASE)
+        or re.search(r'^\s*let me\b', s, flags=re.IGNORECASE)
+        or re.search(r'用户(要求|希望|想让|让我)', s)
+        or re.search(r'请只?回复', s)
+        or re.search(r'^\s*(ok|okay|done|all set|complete|completed|finished)\b[\s.!?]*$', s, flags=re.IGNORECASE)
+        or re.search(r'^\s*(好的|好啦|完成了|已完成|测试完成|测试已完成|可以了|没问题)\s*[！!。\.\s]*$', s)
+    )
+
+
+def _message_text(value) -> str:
+    """Extract plain text from mixed message content payloads."""
+    if isinstance(value, list):
+        parts = []
+        for p in value:
+            if not isinstance(p, dict):
+                continue
+            ptype = str(p.get('type') or '').lower()
+            if ptype in ('', 'text', 'input_text', 'output_text'):
+                parts.append(str(p.get('text') or p.get('content') or ''))
+        return _strip_thinking_markup('\n'.join(parts).strip())
+    return _strip_thinking_markup(str(value or '').strip())
+
+
+def _first_exchange_snippets(messages):
+    """Return (first_user_text, first_assistant_text) snippets for title generation."""
+    user_text = ''
+    asst_text = ''
+    for m in messages or []:
+        if not isinstance(m, dict):
+            continue
+        role = m.get('role')
+        if role == 'user' and not user_text:
+            user_text = _message_text(m.get('content'))
+        elif role == 'assistant' and not asst_text:
+            asst_text = _message_text(m.get('content'))
+        if user_text and asst_text:
+            break
+    return user_text[:500], asst_text[:500]
+
+
+def _is_provisional_title(current_title: str, messages) -> bool:
+    """Heuristic: title equals first-message substring placeholder."""
+    derived = title_from(messages, '') or ''
+    if not derived:
+        return False
+    return (str(current_title or '').strip() == derived[:64])
+
+
+def _title_prompts(user_text: str, assistant_text: str) -> tuple[str, list[str]]:
+    qa = f"User question:\n{user_text[:500]}\n\nAssistant answer:\n{assistant_text[:500]}"
+    prompts = [
+        (
+            "Generate a short session title from this conversation start.\n"
+            "Use BOTH the user's question and the assistant's visible answer.\n"
+            "Return only the title text, 3-8 words, as a topic label.\n"
+            "Do not output a full sentence.\n"
+            "Do not output acknowledgements or completion phrases like OK, done, all set, 测试完成.\n"
+            "Do not describe internal reasoning.\n"
+            "Bad: The user is asking..., OK, 好的，测试完成！\n"
+            "Good: 自动标题生成测试, Clarify Dialog Layout, GitHub Issue Triage"
+        ),
+        (
+            "Rewrite this conversation start as a concise noun-phrase title.\n"
+            "Use the actual topic, not the task outcome.\n"
+            "Return title text only.\n"
+            "Never output acknowledgements, completion status, or meta commentary."
+        ),
+    ]
+    return qa, prompts
+
+
+def _is_minimax_route(provider: str = '', model: str = '', base_url: str = '') -> bool:
+    text = ' '.join([
+        str(provider or '').lower(),
+        str(model or '').lower(),
+        str(base_url or '').lower(),
+    ])
+    return 'minimax' in text or 'minimaxi.com' in text
+
+
+def _title_completion_budget(provider: str = '', model: str = '', base_url: str = '') -> int:
+    if _is_minimax_route(provider, model, base_url):
+        return 384
+    return 160
+
+
+def generate_title_raw_via_aux(
+    user_text: str,
+    assistant_text: str,
+    provider: str = '',
+    model: str = '',
+    base_url: str = '',
+) -> tuple[Optional[str], str]:
+    """Return (raw_text, status) via auxiliary LLM route."""
+    if not user_text or not assistant_text:
+        return None, 'missing_exchange'
+    qa, prompts = _title_prompts(user_text, assistant_text)
+    max_tokens = _title_completion_budget(provider, model, base_url)
+    reasoning_extra = {"reasoning": {"enabled": False}}
+    if _is_minimax_route(provider, model, base_url):
+        reasoning_extra["reasoning_split"] = True
+    try:
+        from agent.auxiliary_client import call_llm
+        for idx, prompt in enumerate(prompts):
+            messages = [
+                {"role": "system", "content": prompt},
+                {"role": "user", "content": qa},
+            ]
+            try:
+                resp = call_llm(
+                    task='title_generation',
+                    provider=provider or None,
+                    model=model or None,
+                    base_url=base_url or None,
+                    messages=messages,
+                    max_tokens=max_tokens,
+                    temperature=0.2,
+                    timeout=15.0,
+                    extra_body=reasoning_extra,
+                )
+                raw = ''
+                try:
+                    raw = resp.choices[0].message.content or ''
+                except Exception:
+                    raw = ''
+                raw = str(raw or '').strip()
+                if raw:
+                    return raw, ('llm_aux' if idx == 0 else 'llm_aux_retry')
+            except Exception as e:
+                logger.debug("Aux title generation attempt %s failed: %s", idx + 1, e)
+        return None, 'llm_error_aux'
+    except Exception as e:
+        logger.debug("Aux title generation failed: %s", e)
+        return None, 'llm_error_aux'
+
+
+def generate_title_raw_via_agent(agent, user_text: str, assistant_text: str) -> tuple[Optional[str], str]:
+    """Return (raw_text, status) via active-agent route."""
+    if not user_text or not assistant_text:
+        return None, 'missing_exchange'
+    if agent is None:
+        return None, 'missing_agent'
+
+    qa, prompts = _title_prompts(user_text, assistant_text)
+    max_tokens = _title_completion_budget(
+        getattr(agent, 'provider', ''),
+        getattr(agent, 'model', ''),
+        getattr(agent, 'base_url', ''),
+    )
+    disabled_reasoning = {"enabled": False}
+    prev_reasoning = getattr(agent, 'reasoning_config', None)
+    try:
+        agent.reasoning_config = disabled_reasoning
+        for idx, prompt in enumerate(prompts):
+            api_messages = [
+                {"role": "system", "content": prompt},
+                {"role": "user", "content": qa},
+            ]
+            try:
+                raw = ""
+                if getattr(agent, 'api_mode', '') == 'codex_responses':
+                    codex_kwargs = agent._build_api_kwargs(api_messages)
+                    codex_kwargs.pop('tools', None)
+                    if 'max_output_tokens' in codex_kwargs:
+                        codex_kwargs['max_output_tokens'] = max_tokens
+                    resp = agent._run_codex_stream(codex_kwargs)
+                    assistant_message, _ = agent._normalize_codex_response(resp)
+                    raw = (assistant_message.content or '') if assistant_message else ''
+                elif getattr(agent, 'api_mode', '') == 'anthropic_messages':
+                    from agent.anthropic_adapter import build_anthropic_kwargs, normalize_anthropic_response
+                    ant_kwargs = build_anthropic_kwargs(
+                        model=agent.model,
+                        messages=api_messages,
+                        tools=None,
+                        max_tokens=max_tokens,
+                        reasoning_config=disabled_reasoning,
+                        is_oauth=getattr(agent, '_is_anthropic_oauth', False),
+                        preserve_dots=agent._anthropic_preserve_dots(),
+                        base_url=getattr(agent, '_anthropic_base_url', None),
+                    )
+                    resp = agent._anthropic_messages_create(ant_kwargs)
+                    assistant_message, _ = normalize_anthropic_response(
+                        resp, strip_tool_prefix=getattr(agent, '_is_anthropic_oauth', False)
+                    )
+                    raw = (assistant_message.content or '') if assistant_message else ''
+                else:
+                    api_kwargs = agent._build_api_kwargs(api_messages)
+                    api_kwargs.pop('tools', None)
+                    api_kwargs['temperature'] = 0.1
+                    api_kwargs['timeout'] = 15.0
+                    if _is_minimax_route(getattr(agent, 'provider', ''), getattr(agent, 'model', ''), getattr(agent, 'base_url', '')):
+                        extra_body = dict(api_kwargs.get('extra_body') or {})
+                        extra_body['reasoning_split'] = True
+                        api_kwargs['extra_body'] = extra_body
+                    if 'max_completion_tokens' in api_kwargs:
+                        api_kwargs['max_completion_tokens'] = max_tokens
+                    else:
+                        api_kwargs['max_tokens'] = max_tokens
+                    resp = agent._ensure_primary_openai_client(reason='title_generation').chat.completions.create(
+                        **api_kwargs,
+                    )
+                    try:
+                        raw = resp.choices[0].message.content or ""
+                    except Exception:
+                        raw = ""
+                raw = str(raw or '').strip()
+                if raw:
+                    return raw, ('llm' if idx == 0 else 'llm_retry')
+            except Exception as e:
+                logger.debug(
+                    "Agent title generation attempt %s failed: provider=%s model=%s error=%s",
+                    idx + 1,
+                    getattr(agent, 'provider', None),
+                    getattr(agent, 'model', None),
+                    e,
+                )
+        return None, 'llm_error'
+    except Exception as e:
+        logger.debug("Agent title generation failed: %s", e)
+        return None, 'llm_error'
+    finally:
+        agent.reasoning_config = prev_reasoning
+
+
+def _generate_llm_session_title_for_agent(agent, user_text: str, assistant_text: str) -> tuple[Optional[str], str, str]:
+    """Generate a title via active-agent route, then sanitize/validate result."""
+    raw, status = generate_title_raw_via_agent(agent, user_text, assistant_text)
+    if not raw:
+        return None, status, ''
+    title = _sanitize_generated_title(raw)
+    if title:
+        return title, status, ''
+    return None, 'llm_invalid', str(raw)[:120]
+
+
+def _generate_llm_session_title_via_aux(user_text: str, assistant_text: str, agent=None) -> tuple[Optional[str], str, str]:
+    """Generate a title via dedicated auxiliary LLM route, then sanitize/validate result."""
+    raw, status = generate_title_raw_via_aux(
+        user_text,
+        assistant_text,
+        provider=getattr(agent, 'provider', '') if agent else '',
+        model=getattr(agent, 'model', '') if agent else '',
+        base_url=getattr(agent, 'base_url', '') if agent else '',
+    )
+    if not raw:
+        return None, status, ''
+    title = _sanitize_generated_title(raw)
+    if title:
+        return title, status, ''
+    return None, 'llm_invalid_aux', str(raw)[:120]
+
+
+def _put_title_status(put_event, session_id: str, status: str, reason: str = '', title: str = '', raw_preview: str = '') -> None:
+    payload = {'session_id': session_id, 'status': status}
+    if reason:
+        payload['reason'] = reason
+    if title:
+        payload['title'] = title
+    if raw_preview:
+        payload['raw_preview'] = raw_preview
+    put_event('title_status', payload)
+    logger.info(
+        "title_status session=%s status=%s reason=%s title=%r raw_preview=%r",
+        session_id,
+        status,
+        reason or '-',
+        title or '',
+        (raw_preview or '')[:120],
+    )
+
+
+def _fallback_title_from_exchange(user_text: str, assistant_text: str) -> Optional[str]:
+    """Generate a readable local fallback title when LLM title generation fails."""
+    user_text = (user_text or '').strip()
+    assistant_text = _strip_thinking_markup(assistant_text or '').strip()
+    if not user_text:
+        return None
+    user_text = re.sub(r'^\[Workspace:[^\]]+\]\s*', '', user_text)
+    user_text = re.sub(r'\s+', ' ', user_text).strip()
+    assistant_text = re.sub(r'\s+', ' ', assistant_text).strip()
+    combined = f"{user_text} {assistant_text}".strip().lower()
+    combined_raw = f"{user_text} {assistant_text}".strip()
+
+    def _extract_named_topic(text: str) -> str:
+        m = re.search(r'《([^》]{2,24})》', text)
+        if m:
+            return (m.group(1) or '').strip()
+        m = re.search(r'"([^"\n]{2,24})"', text)
+        if m:
+            return (m.group(1) or '').strip()
+        m = re.search(r'“([^”\n]{2,24})”', text)
+        if m:
+            return (m.group(1) or '').strip()
+        return ''
+
+    topic_name = _extract_named_topic(combined_raw)
+    if topic_name:
+        if any(k in combined for k in ('时间', 'time', '安排', '效率', '怎么办', '健身', '唱歌', '写毛笔', '不够用了')):
+            return f'{topic_name}与时间管理'
+        if any(k in combined for k in ('hermes', 'codex', 'ai')):
+            return f'{topic_name}与AI效率'
+        return f'{topic_name}讨论'
+
+    if any(k in combined for k in ('title', '标题')) and any(k in combined for k in ('summary', 'summar', '摘要', '短标题')):
+        if any(k in combined for k in ('test', '测试', 'ok', '回复ok')):
+            return '会话标题自动摘要测试'
+        return '会话标题自动摘要'
+    if any(k in combined for k in ('clarify', '澄清')) and any(k in combined for k in ('dialog', 'card', '对话', '卡片')):
+        return 'Clarify 对话卡片'
+    if any(k in combined for k in ('issue', 'github', 'pr')) and any(k in combined for k in ('triage', 'bug', 'review', '问题')):
+        return 'GitHub Issue Triage'
+
+    head = re.split(r'[。！？.!?\n]', user_text)[0].strip()
+    if not head:
+        return None
+
+    stop_cjk = {
+        '我们', '看看', '一下', '这个', '标题', '是否', '可以', '用户', '理解', '这里', '测试', '一下',
+        '你只', '需要', '回复', '就可', '可以', '不需', '需要做', '什么', '自动', '成用户', '短标题',
+    }
+    stop_en = {
+        'the', 'this', 'that', 'with', 'from', 'into', 'just', 'reply', 'please',
+        'need', 'needs', 'want', 'wants', 'user', 'assistant', 'could', 'would',
+        'should', 'about', 'there', 'here', 'test', 'testing', 'title', 'summary',
+    }
+    tokens = re.findall(r'[\u4e00-\u9fff]{2,6}|[A-Za-z0-9][A-Za-z0-9_./+-]*', head)
+    if not tokens:
+        return head[:64]
+
+    picked = []
+    for tok in tokens:
+        lower_tok = tok.lower()
+        if re.search(r'[\u4e00-\u9fff]', tok):
+            if tok in stop_cjk:
+                continue
+        else:
+            if lower_tok in stop_en or len(lower_tok) < 3:
+                continue
+        if tok not in picked:
+            picked.append(tok)
+        if len(picked) >= 4:
+            break
+
+    if picked:
+        if any(re.search(r'[\u4e00-\u9fff]', t) for t in picked):
+            return ''.join(picked)[:20]
+        return ' '.join(picked)[:60]
+    return head[:24]
+
+
+def _run_background_title_update(session_id: str, user_text: str, assistant_text: str, placeholder_title: str, put_event, agent=None):
+    """Generate and publish a better title after `done`, then end the stream."""
+    try:
+        try:
+            s = get_session(session_id)
+        except KeyError:
+            _put_title_status(put_event, session_id, 'skipped', 'missing_session')
+            return
+        # Allow self-heal when a previously generated title leaked thinking text.
+        _invalid_existing = _looks_invalid_generated_title(s.title)
+        if getattr(s, 'llm_title_generated', False) and not _invalid_existing:
+            _put_title_status(put_event, session_id, 'skipped', 'already_generated', str(s.title or ''))
+            return
+        current = str(s.title or '').strip()
+        still_auto = (
+            current == placeholder_title
+            or current in ('Untitled', 'New Chat', '')
+            or _is_provisional_title(current, s.messages)
+            or _invalid_existing
+        )
+        if not still_auto:
+            _put_title_status(put_event, session_id, 'skipped', 'manual_title', current)
+            return
+        # Prefer the active session model when available so title generation
+        # matches the user's chosen runtime and can use provider-specific fixes.
+        if agent:
+            next_title, llm_status, raw_preview = _generate_llm_session_title_for_agent(agent, user_text, assistant_text)
+            if not next_title and llm_status in ('llm_error', 'llm_invalid'):
+                next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, agent=agent)
+        else:
+            next_title, llm_status, raw_preview = _generate_llm_session_title_via_aux(user_text, assistant_text, agent=agent)
+        source = llm_status
+        if not next_title:
+            next_title = _fallback_title_from_exchange(user_text, assistant_text)
+            if next_title:
+                logger.debug("Using local fallback for session title generation")
+                source = 'fallback'
+        if next_title and next_title != current:
+            s.title = next_title
+            s.llm_title_generated = True
+            # Keep chronological ordering stable in the sidebar.
+            s.save(touch_updated_at=False)
+            if source == 'fallback':
+                _put_title_status(put_event, session_id, source, 'local_summary', s.title, raw_preview)
+            else:
+                _put_title_status(put_event, session_id, source, llm_status, s.title, raw_preview)
+            put_event('title', {'session_id': s.session_id, 'title': s.title})
+        else:
+            _put_title_status(put_event, session_id, 'skipped', source or 'unchanged', current, raw_preview)
+    finally:
+        put_event('stream_end', {'session_id': session_id})
+
+
 def _sanitize_messages_for_api(messages):
     """Return a deep copy of messages with only API-safe fields.
 
@@ -543,6 +981,17 @@ def _run_agent_streaming(session_id, msg_text, model, workspace, stream_id, atta
             # Only auto-generate title when still default; preserves user renames
             if s.title == 'Untitled' or s.title == 'New Chat' or not s.title:
                 s.title = title_from(s.messages, s.title)
+            _looks_default = (s.title == 'Untitled' or s.title == 'New Chat' or not s.title)
+            _looks_provisional = _is_provisional_title(s.title, s.messages)
+            _invalid_existing_title = _looks_invalid_generated_title(s.title)
+            _should_bg_title = (
+                (_looks_default or _looks_provisional or _invalid_existing_title)
+                and (not getattr(s, 'llm_title_generated', False) or _invalid_existing_title)
+            )
+            _u0 = ''
+            _a0 = ''
+            if _should_bg_title:
+                _u0, _a0 = _first_exchange_snippets(s.messages)
             # Read token/cost usage from the agent object (if available)
             input_tokens = getattr(agent, 'session_prompt_tokens', 0) or 0
             output_tokens = getattr(agent, 'session_completion_tokens', 0) or 0
@@ -657,6 +1106,14 @@ def _run_agent_streaming(session_id, msg_text, model, workspace, stream_id, atta
                         break
             raw_session = s.compact() | {'messages': s.messages, 'tool_calls': tool_calls}
             put('done', {'session': redact_session_data(raw_session), 'usage': usage})
+            if _should_bg_title and _u0 and _a0:
+                threading.Thread(
+                    target=_run_background_title_update,
+                    args=(s.session_id, _u0, _a0, str(s.title or '').strip(), put, agent),
+                    daemon=True,
+                ).start()
+            else:
+                put('stream_end', {'session_id': s.session_id})
         finally:
             # Unregister the gateway approval callback and unblock any threads
             # still waiting on approval (e.g. stream cancelled mid-approval).
diff --git a/static/index.html b/static/index.html
index 4438fd8..d0b90fe 100644
--- a/static/index.html
+++ b/static/index.html
@@ -553,7 +553,7 @@
                 <div class="settings-section-title">System</div>
                 <div class="settings-section-meta">Instance version and access controls.</div>
               </div>
-              <span class="settings-version-badge">v0.50.55</span>
+              <span class="settings-version-badge">v0.50.57</span>
             </div>
             <div class="settings-field" style="border-top:1px solid var(--border);padding-top:12px;margin-top:8px">
               <label for="settingsPassword" data-i18n="settings_label_password">Access Password</label>
diff --git a/static/messages.js b/static/messages.js
index 8e37699..4696119 100644
--- a/static/messages.js
+++ b/static/messages.js
@@ -315,7 +315,12 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
       const d=JSON.parse(e.data);
       if(d.name==='clarify') return;
       const tc={name:d.name, preview:d.preview||'', args:d.args||{}, snippet:'', done:false, tid:d.tid||`live-${Date.now()}-${Math.random().toString(36).slice(2,8)}`};
-      if(!Array.isArray(INFLIGHT[activeSid].toolCalls)) INFLIGHT[activeSid].toolCalls=[];
+      const inflight = INFLIGHT[activeSid] || (INFLIGHT[activeSid] = {
+        messages:[...S.messages],
+        uploaded:[],
+        toolCalls:[]
+      });
+      if(!Array.isArray(inflight.toolCalls)) inflight.toolCalls=[];
       INFLIGHT[activeSid].toolCalls.push(tc);
       S.toolCalls=INFLIGHT[activeSid].toolCalls;
       persistInflightState();
@@ -373,8 +378,40 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
       sendBrowserNotification('Clarification needed',d.question||'Tool clarification needed');
     });
 
+    source.addEventListener('title',e=>{
+      let d={};
+      try{ d=JSON.parse(e.data||'{}'); }catch(_){}
+      if((d.session_id||activeSid)!==activeSid) return;
+      const newTitle=String(d.title||'').trim();
+      if(!newTitle) return;
+      if(S.session&&S.session.session_id===activeSid){
+        S.session.title=newTitle;
+        syncTopbar();
+      }
+      if(typeof _allSessions!=='undefined'&&Array.isArray(_allSessions)){
+        const row=_allSessions.find(s=>s&&s.session_id===activeSid);
+        if(row) row.title=newTitle;
+      }
+      if(typeof renderSessionListFromCache==='function') renderSessionListFromCache();
+      else if(typeof renderSessionList==='function') renderSessionList();
+    });
+
+    source.addEventListener('title_status',e=>{
+      let d={};
+      try{ d=JSON.parse(e.data||'{}'); }catch(_){}
+      if((d.session_id||activeSid)!==activeSid) return;
+      try{
+        console.info('[title]', {
+          status:String(d.status||''),
+          reason:String(d.reason||''),
+          title:String(d.title||''),
+          raw_preview:String(d.raw_preview||''),
+          session_id:String(d.session_id||activeSid)
+        });
+      }catch(_){}
+    });
+
     source.addEventListener('done',e=>{
-      source.close();
       const d=JSON.parse(e.data);
       delete INFLIGHT[activeSid];
       clearInflight();clearInflightState(activeSid);
@@ -416,6 +453,14 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
       sendBrowserNotification('Response complete',assistantText?assistantText.slice(0,100):'Task finished');
     });
 
+    source.addEventListener('stream_end',e=>{
+      try{
+        const d=JSON.parse(e.data||'{}');
+        if((d.session_id||activeSid)!==activeSid) return;
+      }catch(_){}
+      source.close();
+    });
+
     source.addEventListener('compressed',e=>{
       // Context was auto-compressed during this turn -- show a system message
       if(!S.session||S.session.session_id!==activeSid) return;
@@ -526,7 +571,36 @@ function attachLiveStream(activeSid, streamId, uploaded=[], options={}){
     if(!S.session||!INFLIGHT[S.session.session_id]){setBusy(false);setComposerStatus('');}
   }
 
-  _wireSSE(new EventSource(new URL(`/api/chat/stream?stream_id=${encodeURIComponent(streamId)}`,location.origin).href,{withCredentials:true}));
+  (async()=>{
+    // Reattach path can carry stale stream ids after server restart; preflight
+    // status avoids opening a dead SSE URL that will 404 in the console.
+    if(reconnecting){
+      try{
+        const st=await api(`/api/chat/stream/status?stream_id=${encodeURIComponent(streamId)}`);
+        if(!st.active){
+          delete INFLIGHT[activeSid];
+          clearInflight();
+          clearInflightState(activeSid);
+          stopApprovalPolling();
+          stopClarifyPolling();
+          if(!_approvalSessionId||_approvalSessionId===activeSid) hideApprovalCard(true);
+          if(!_clarifySessionId||_clarifySessionId===activeSid) hideClarifyCard(true);
+          if(S.session&&S.session.session_id===activeSid){
+            S.activeStreamId=null;
+            const _cbe=$('btnCancel');if(_cbe)_cbe.style.display='none';
+            clearLiveToolCards();
+            removeThinking();
+            setBusy(false);
+            setComposerStatus('');
+            renderMessages();
+            renderSessionList();
+          }
+          return;
+        }
+      }catch(_){}
+    }
+    _wireSSE(new EventSource(new URL(`/api/chat/stream?stream_id=${encodeURIComponent(streamId)}`,location.origin).href,{withCredentials:true}));
+  })();
 
 }
 
diff --git a/tests/test_issue487b.py b/tests/test_issue487b.py
index 097f043..18e4374 100644
--- a/tests/test_issue487b.py
+++ b/tests/test_issue487b.py
@@ -1,4 +1,4 @@
-"""
+r"""
 Regression test for image src URL corruption by the autolink pass.
 
 Bug: the _al_stash before the autolink pass only stashed <a> tags.
diff --git a/tests/test_sprint41.py b/tests/test_sprint41.py
index 5e0e5f5..19e112e 100644
--- a/tests/test_sprint41.py
+++ b/tests/test_sprint41.py
@@ -18,6 +18,7 @@ import unittest
 REPO_ROOT = pathlib.Path(__file__).parent.parent
 CSS = (REPO_ROOT / "static" / "style.css").read_text()
 HTML = (REPO_ROOT / "static" / "index.html").read_text()
+MESSAGES_JS = (REPO_ROOT / "static" / "messages.js").read_text()
 STREAMING_PY = (REPO_ROOT / "api" / "streaming.py").read_text()
 
 
@@ -125,5 +126,93 @@ class TestWorkspacePanelButtons(unittest.TestCase):
                       "mobile-close-btn must have aria-label for accessibility")
 
 
+class TestIssue495TitleStreaming(unittest.TestCase):
+    """Regression checks for issue #495 title SSE behavior."""
+
+    def test_streaming_has_llm_title_helper(self):
+        self.assertIn(
+            "def _generate_llm_session_title_for_agent(",
+            STREAMING_PY,
+            "streaming.py should define an agent-backed LLM title helper for session titles",
+        )
+
+    def test_streaming_rejects_generic_completion_titles(self):
+        self.assertIn(
+            "测试完成",
+            STREAMING_PY,
+            "streaming.py should reject generic completion phrases as session titles",
+        )
+        self.assertIn(
+            "all set",
+            STREAMING_PY,
+            "streaming.py should reject generic English completion phrases as session titles",
+        )
+
+    def test_streaming_uses_reasoning_split_for_minimax_titles(self):
+        self.assertIn(
+            "reasoning_split",
+            STREAMING_PY,
+            "streaming.py should request MiniMax title calls with reasoning_split so final text is separated from thinking",
+        )
+
+    def test_streaming_emits_title_sse_event(self):
+        self.assertIn(
+            "put_event('title', {'session_id': s.session_id, 'title': s.title})",
+            STREAMING_PY,
+            "streaming.py should emit a title SSE event when title is updated",
+        )
+
+    def test_streaming_emits_title_status_sse_event(self):
+        self.assertIn(
+            "put_event('title_status', payload)",
+            STREAMING_PY,
+            "streaming.py should emit a title_status SSE event for title generation diagnostics",
+        )
+
+    def test_streaming_emits_stream_end_event(self):
+        self.assertIn(
+            "put_event('stream_end', {'session_id': session_id})",
+            STREAMING_PY,
+            "background title path should end the SSE stream with stream_end",
+        )
+
+    def test_frontend_listens_for_title_event(self):
+        self.assertIn(
+            "addEventListener('title'",
+            MESSAGES_JS,
+            "messages.js should listen for title SSE events",
+        )
+
+    def test_frontend_listens_for_title_status_event(self):
+        self.assertIn(
+            "addEventListener('title_status'",
+            MESSAGES_JS,
+            "messages.js should listen for title_status SSE events",
+        )
+        self.assertIn(
+            "console.info('[title]'",
+            MESSAGES_JS,
+            "messages.js should log title generation diagnostics to the browser console",
+        )
+
+    def test_frontend_refreshes_title_ui_after_title_event(self):
+        self.assertIn(
+            "syncTopbar()",
+            MESSAGES_JS,
+            "messages.js title listener should sync top bar title",
+        )
+        self.assertTrue(
+            ("renderSessionListFromCache()" in MESSAGES_JS) or ("renderSessionList()" in MESSAGES_JS),
+            "messages.js title listener should refresh session list UI",
+        )
+
+    def test_frontend_waits_for_stream_end_before_closing(self):
+        self.assertIn(
+            "addEventListener('stream_end'",
+            MESSAGES_JS,
+            "messages.js should close SSE connection on stream_end (not immediately on done)",
+        )
+
+
 if __name__ == "__main__":
     unittest.main()