coderouter-cli 2.3.0a4__py3-none-any.whl → 2.5.0__py3-none-any.whl

coderouter/ingress/launcher_routes.py

@@ -0,0 +1,1176 @@
+"""Launcher routes — ``GET /launcher`` + ``/api/launcher/*``.
+
+llama.cpp / vllm プロセス管理 UI。
+
+設計方針:
+- ダッシュボードと同じ "1ファイル完結" スタイル (Tailwind CDN + inline JS)
+- プロセスレジストリは app.state.launcher に持たせる (再起動で消えるが意図通り)
+- option_profiles は providers.yaml の launcher: セクションで管理 → コード変更不要で拡張可
+- 複数プロセスの同時起動に対応 (UUID ベースの ID 管理)
+- llama.cpp / vllm どちらも同じ key-value args スキーマで統一
+
+エンドポイント:
+  GET  /launcher                   → HTML UI
+  GET  /api/launcher/models        → model_dirs をスキャンしてリスト返却
+  GET  /api/launcher/option-profiles → providers.yaml の option_profiles を返却
+  GET  /api/launcher/processes     → 起動中・停止済みプロセス一覧
+  POST /api/launcher/start         → プロセス起動
+  POST /api/launcher/stop/{id}     → プロセス停止 (SIGTERM → SIGKILL)
+  DELETE /api/launcher/processes/{id} → レジストリから削除 (停止済みのみ)
+  GET  /api/launcher/logs/{id}     → ログ最新 N 行
+"""
+
+from __future__ import annotations
+
+import asyncio
+import contextlib
+import os
+import platform
+import shlex
+import shutil
+import subprocess
+import uuid
+from collections import deque
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any
+
+from fastapi import APIRouter, HTTPException, Request
+from fastapi.responses import HTMLResponse
+from pydantic import BaseModel
+
+router = APIRouter()
+
+# 背景タスクへの強参照を保持する (create_task の戻り値が GC されるのを防ぐ)
+_background_tasks: set[asyncio.Task[Any]] = set()
+
+# ---------------------------------------------------------------------------
+# Model file extensions to scan
+# ---------------------------------------------------------------------------
+
+_MODEL_EXTS = {".gguf", ".ggml", ".safetensors", ".bin", ".pt", ".pth"}
+
+
+# ---------------------------------------------------------------------------
+# ManagedProcess — 起動したプロセスの状態を保持
+# ---------------------------------------------------------------------------
+
+
+@dataclass
+class ManagedProcess:
+    """Running or stopped backend process entry."""
+
+    id: str
+    name: str
+    backend: str         # "llama.cpp" | "vllm"
+    model_path: str
+    port: int
+    options: dict[str, Any]
+    extra_args: str
+    status: str = "starting"   # "starting" | "running" | "stopped" | "error"
+    pid: int | None = None
+    returncode: int | None = None
+    log_tail: deque = field(default_factory=lambda: deque(maxlen=200))
+    # asyncio subprocess handle — not serialised
+    _proc: Any = field(default=None, repr=False, compare=False)
+
+
+# ---------------------------------------------------------------------------
+# LauncherRegistry — app.state に格納するレジストリ
+# ---------------------------------------------------------------------------
+
+
+class LauncherRegistry:
+    """In-process registry for ManagedProcess instances."""
+
+    def __init__(self) -> None:
+        self._procs: dict[str, ManagedProcess] = {}
+
+    def get(self, proc_id: str) -> ManagedProcess:
+        try:
+            return self._procs[proc_id]
+        except KeyError:
+            raise KeyError(proc_id) from None
+
+    def add(self, proc: ManagedProcess) -> None:
+        self._procs[proc.id] = proc
+
+    def remove(self, proc_id: str) -> None:
+        del self._procs[proc_id]
+
+    def all(self) -> list[ManagedProcess]:
+        return list(self._procs.values())
+
+
+def _registry(request: Request) -> LauncherRegistry:
+    """Get or create the LauncherRegistry on app.state."""
+    if not hasattr(request.app.state, "launcher"):
+        request.app.state.launcher = LauncherRegistry()
+    return request.app.state.launcher
+
+
+# ---------------------------------------------------------------------------
+# Model scanning
+# ---------------------------------------------------------------------------
+
+
+def _scan_models(model_dirs: list[str]) -> list[dict[str, Any]]:
+    """Walk model_dirs and return metadata for each discovered model file."""
+    found: list[dict[str, Any]] = []
+    for raw in model_dirs:
+        base = Path(raw).expanduser().resolve()
+        if not base.exists():
+            continue
+        for p in sorted(base.rglob("*")):
+            if not p.is_file():
+                continue
+            if p.suffix.lower() not in _MODEL_EXTS:
+                continue
+            try:
+                size = p.stat().st_size
+            except OSError:
+                continue
+            found.append(
+                {
+                    "path": str(p),
+                    "name": p.name,
+                    "dir": str(p.parent),
+                    "size_gb": round(size / (1024**3), 2),
+                    "ext": p.suffix.lower(),
+                }
+            )
+    return found
+
+
+# ---------------------------------------------------------------------------
+# Command builder
+# ---------------------------------------------------------------------------
+
+
+_BACKEND_DEFAULTS: dict[str, str] = {
+    "llama.cpp": "llama-server",
+    "vllm": "python",
+}
+
+
+def _resolve_binary(backend: str, configured: str | None) -> str:
+    """Return the executable to use, expanding ~ and env vars."""
+    raw = configured or _BACKEND_DEFAULTS.get(backend, backend)
+    return str(Path(raw).expanduser())
+
+
+def _resolve_backends_sync(
+    backends_cfg: dict[str, Any] | None,
+) -> dict[str, dict[str, Any]]:
+    """Resolve binary paths and check availability for every backend.
+
+    Performs blocking filesystem I/O (``is_file`` / ``shutil.which``),
+    so async-route callers must invoke it via ``asyncio.to_thread``.
+    """
+    result: dict[str, dict[str, Any]] = {}
+    for backend, default_bin in _BACKEND_DEFAULTS.items():
+        configured: str | None = None
+        if backends_cfg:
+            bc = backends_cfg.get(backend)
+            if bc and bc.binary:
+                configured = bc.binary
+        resolved = _resolve_binary(backend, configured)
+        expanded = str(Path(resolved).expanduser())
+        found = (
+            Path(expanded).is_file()               # フルパス指定でファイルが存在
+            or shutil.which(expanded) is not None   # PATH から解決可能
+        )
+        result[backend] = {
+            "resolved": resolved,
+            "configured": configured or "",
+            "default": default_bin,
+            "is_custom": configured is not None,
+            "found": found,
+        }
+    return result
+
+
+# ---------------------------------------------------------------------------
+# Hardware detection + model recommendation (luna-go /models 互換の発想)
+# ---------------------------------------------------------------------------
+
+
+def _detect_hardware() -> dict[str, Any]:
+    """ハードウェアを best-effort で検出する。
+
+    ブロッキング I/O (sysctl / nvidia-smi) を含むため、async ルートからは
+    ``asyncio.to_thread`` 経由で呼ぶこと。
+    """
+    cpu = os.cpu_count() or 4
+    ram_gb = 0.0
+    with contextlib.suppress(ValueError, OSError, AttributeError):
+        ram_gb = (os.sysconf("SC_PHYS_PAGES")
+                  * os.sysconf("SC_PAGE_SIZE") / (1024 ** 3))
+    if ram_gb <= 0:
+        try:
+            out = subprocess.run(["sysctl", "-n", "hw.memsize"],
+                                 capture_output=True, text=True, timeout=3)
+            ram_gb = int(out.stdout.strip()) / (1024 ** 3)
+        except (ValueError, OSError, subprocess.SubprocessError):
+            pass
+    gpu, vram_gb = "cpu", 0.0
+    if platform.system() == "Darwin" and platform.machine() == "arm64":
+        gpu, vram_gb = "metal", ram_gb            # ユニファイドメモリ
+    elif shutil.which("nvidia-smi"):
+        try:
+            out = subprocess.run(
+                ["nvidia-smi", "--query-gpu=memory.total",
+                 "--format=csv,noheader,nounits"],
+                capture_output=True, text=True, timeout=5)
+            mb = max((int(x) for x in out.stdout.split() if x.strip().isdigit()),
+                     default=0)
+            if mb > 0:
+                gpu, vram_gb = "cuda", mb / 1024
+        except (ValueError, OSError, subprocess.SubprocessError):
+            pass
+    return {"ram_gb": round(ram_gb, 1), "vram_gb": round(vram_gb, 1),
+            "gpu": gpu, "cpu_count": cpu}
+
+
+def _usable_memory_gb(hw: dict[str, Any]) -> float:
+    """モデルの重み + KV キャッシュに使えるメモリ量。"""
+    if hw.get("gpu") == "cuda":
+        return float(hw.get("vram_gb", 0.0))
+    return float(hw.get("ram_gb", 0.0))          # metal (ユニファイド) / cpu
+
+
+def _model_recommendation(size_gb: float, hw: dict[str, Any]) -> dict[str, str]:
+    """モデル単位のメモリ適合判定 (luna-go /models 相当)。
+
+    level: "ok" (推奨) | "warn" (メモリ厳しい) | "unknown"
+    """
+    usable = _usable_memory_gb(hw)
+    if usable <= 0 or size_gb <= 0:
+        return {"level": "unknown", "label": "—"}
+    if size_gb * 1.2 + 2.0 <= usable:
+        return {"level": "ok", "label": "推奨"}
+    return {"level": "warn", "label": "メモリ厳しい"}
+
+
+def _suggest_launch_flags(size_gb: float, hw: dict[str, Any]) -> str:
+    """選択モデル + ハードから -ngl / --ctx-size / --threads を提案する。
+
+    あくまで目安。他プロセスのメモリ使用や量子化方式までは考慮しない。
+    """
+    threads = max(1, int(hw.get("cpu_count", 4)) - 2)
+    usable = _usable_memory_gb(hw)
+    weights = size_gb * 1.15                       # 重み + オーバーヘッド概算
+    if hw.get("gpu") == "cpu":
+        ngl = 0
+    elif usable >= weights + 1.0:
+        ngl = 99                                   # 全レイヤー GPU に載る
+    elif usable > 1.5:
+        ngl = max(0, min(99, int(99 * (usable - 0.7) / max(weights, 0.1))))
+    else:
+        ngl = 0
+    headroom = usable - weights - 1.0
+    if headroom >= 8:
+        ctx = 32768
+    elif headroom >= 4:
+        ctx = 16384
+    elif headroom >= 2:
+        ctx = 8192
+    else:
+        ctx = 4096
+    return f"-ngl {ngl} --ctx-size {ctx} --threads {threads}"
+
+
+def _model_size_gb(path: str) -> float:
+    """モデルファイルのサイズ (GB)。失敗時は 0.0 (ブロッキング — to_thread 推奨)。"""
+    try:
+        return Path(path).expanduser().stat().st_size / (1024 ** 3)
+    except OSError:
+        return 0.0
+
+
+def _build_cmd(
+    backend: str,
+    model_path: str,
+    port: int,
+    options: dict[str, Any],
+    extra_args: str,
+    binary: str | None = None,
+) -> list[str]:
+    """Build the CLI command list for the given backend and options.
+
+    ``binary`` overrides the default executable (``llama-server`` /
+    ``python``).  When None, the default is used and PATH resolution
+    is left to the OS.
+    """
+    exe = _resolve_binary(backend, binary)
+
+    if backend == "llama.cpp":
+        cmd: list[str] = [exe, "-m", model_path, "--port", str(port)]
+    elif backend == "vllm":
+        cmd = [
+            exe, "-m", "vllm.entrypoints.openai.api_server",
+            "--model", model_path,
+            "--port", str(port),
+        ]
+    else:
+        raise ValueError(f"Unknown backend: {backend!r}. Expected 'llama.cpp' or 'vllm'.")
+
+    for flag, val in options.items():
+        if isinstance(val, bool):
+            if val:
+                cmd.append(flag)
+        else:
+            cmd.extend([flag, str(val)])
+
+    if extra_args.strip():
+        cmd.extend(shlex.split(extra_args))
+
+    return cmd
+
+
+# ---------------------------------------------------------------------------
+# Log reader background task
+# ---------------------------------------------------------------------------
+
+
+async def _tail_logs(proc: ManagedProcess) -> None:
+    """Read stdout+stderr into proc.log_tail until the process exits."""
+    p = proc._proc
+    if p is None:
+        return
+
+    async def _drain(stream: asyncio.StreamReader | None) -> None:
+        if stream is None:
+            return
+        while True:
+            line = await stream.readline()
+            if not line:
+                break
+            proc.log_tail.append(line.decode(errors="replace").rstrip())
+
+    await asyncio.gather(_drain(p.stdout), _drain(p.stderr))
+    await p.wait()
+    proc.returncode = p.returncode
+    proc.pid = None
+    proc.status = "stopped" if (p.returncode or 0) == 0 else "error"
+    proc.log_tail.append(
+        f"[launcher] process exited with code {p.returncode}"
+    )
+
+
+async def shutdown_launcher(app: Any) -> None:
+    """Terminate all managed child processes on CodeRouter shutdown.
+
+    Called from the FastAPI lifespan so that llama.cpp / vllm processes
+    started via the Launcher are not left as orphans when CodeRouter exits.
+    """
+    reg = getattr(app.state, "launcher", None)
+    if reg is None:
+        return
+    procs = reg.all()
+    for proc in procs:
+        p = proc._proc
+        if p is not None and p.returncode is None:
+            with contextlib.suppress(Exception):
+                p.terminate()
+    for proc in procs:
+        p = proc._proc
+        if p is None or p.returncode is not None:
+            continue
+        try:
+            await asyncio.wait_for(p.wait(), timeout=5.0)
+        except TimeoutError:
+            with contextlib.suppress(Exception):
+                p.kill()
+        except Exception:
+            pass
+
+
+# ---------------------------------------------------------------------------
+# Pydantic request models
+# ---------------------------------------------------------------------------
+
+
+class StartRequest(BaseModel):
+    name: str
+    backend: str
+    model_path: str
+    port: int
+    options: dict[str, Any] = {}
+    extra_args: str = ""
+
+
+# ---------------------------------------------------------------------------
+# API routes
+# ---------------------------------------------------------------------------
+
+
+@router.get("/api/launcher/models")
+async def api_models(request: Request) -> dict[str, Any]:
+    """Scan model_dirs and return discovered model files."""
+    cfg = request.app.state.config
+    launcher_cfg = getattr(cfg, "launcher", None)
+    model_dirs: list[str] = launcher_cfg.model_dirs if launcher_cfg else []
+    # rglob / stat はブロッキング I/O。イベントループ(= プロキシ全体)を
+    # 止めないよう別スレッドへ退避する。
+    models = await asyncio.to_thread(_scan_models, model_dirs)
+    hw = await asyncio.to_thread(_detect_hardware)
+    for m in models:
+        m["recommendation"] = _model_recommendation(m.get("size_gb", 0.0), hw)
+    return {
+        "models": models,
+        "model_dirs": model_dirs,
+        "hardware": hw,
+    }
+
+
+@router.get("/api/launcher/option-profiles")
+async def api_option_profiles(request: Request) -> dict[str, Any]:
+    """Return option_profiles from providers.yaml launcher config."""
+    cfg = request.app.state.config
+    launcher_cfg = getattr(cfg, "launcher", None)
+    if not launcher_cfg:
+        return {"profiles": {}, "_note": "launcher: block not found in providers.yaml"}
+    if not launcher_cfg.option_profiles:
+        return {"profiles": {}, "_note": "option_profiles is empty — add option_profiles: under launcher: in providers.yaml"}
+    result: dict[str, list[dict]] = {}
+    for backend, profiles in launcher_cfg.option_profiles.items():
+        result[backend] = [{"name": p.name, "args": p.args} for p in profiles]
+    return {"profiles": result}
+
+
+@router.get("/api/launcher/config-debug")
+async def api_launcher_config_debug(request: Request) -> dict[str, Any]:
+    """Return the effective launcher config for troubleshooting."""
+    cfg = request.app.state.config
+    launcher_cfg = getattr(cfg, "launcher", None)
+    if not launcher_cfg:
+        return {"launcher": None, "message": "launcher: block not found in providers.yaml"}
+    return {
+        "launcher": {
+            "model_dirs": launcher_cfg.model_dirs,
+            "backends": {k: {"binary": v.binary} for k, v in launcher_cfg.backends.items()},
+            "option_profiles": {
+                k: [p.name for p in v]
+                for k, v in launcher_cfg.option_profiles.items()
+            },
+        },
+    }
+
+
+@router.get("/api/launcher/processes")
+async def api_processes(request: Request) -> dict[str, Any]:
+    """List all managed processes."""
+    reg = _registry(request)
+    return {
+        "processes": [
+            {
+                "id": p.id,
+                "name": p.name,
+                "backend": p.backend,
+                "model_path": p.model_path,
+                "port": p.port,
+                "status": p.status,
+                "pid": p.pid,
+                "returncode": p.returncode,
+            }
+            for p in reg.all()
+        ]
+    }
+
+
+@router.get("/api/launcher/backends")
+async def api_backends(request: Request) -> dict[str, Any]:
+    """Return resolved binary paths for each backend.
+
+    Used by the UI to display which executable will be invoked.
+    Shows configured path (from providers.yaml) or the PATH default.
+    """
+    cfg = request.app.state.config
+    launcher_cfg = getattr(cfg, "launcher", None)
+    backends_cfg = (
+        launcher_cfg.backends
+        if (launcher_cfg and launcher_cfg.backends)
+        else None
+    )
+    # is_file / shutil.which はブロッキング I/O。別スレッドへ退避する。
+    result = await asyncio.to_thread(_resolve_backends_sync, backends_cfg)
+    return {"backends": result}
+
+
+@router.post("/api/launcher/start")
+async def api_start(req: StartRequest, request: Request) -> dict[str, Any]:
+    """Start a new backend process."""
+    # Resolve binary path from providers.yaml launcher.backends
+    cfg = request.app.state.config
+    launcher_cfg = getattr(cfg, "launcher", None)
+    configured_binary: str | None = None
+    if launcher_cfg and launcher_cfg.backends:
+        bc = launcher_cfg.backends.get(req.backend)
+        if bc and bc.binary:
+            configured_binary = bc.binary
+
+    try:
+        cmd = _build_cmd(
+            req.backend, req.model_path, req.port,
+            req.options, req.extra_args,
+            binary=configured_binary,
+        )
+    except ValueError as exc:
+        raise HTTPException(status_code=400, detail=str(exc)) from exc
+
+    proc_id = uuid.uuid4().hex[:8]
+    proc = ManagedProcess(
+        id=proc_id,
+        name=req.name,
+        backend=req.backend,
+        model_path=req.model_path,
+        port=req.port,
+        options=req.options,
+        extra_args=req.extra_args,
+        status="starting",
+    )
+    proc.log_tail.append(f"[launcher] cmd: {' '.join(cmd)}")
+
+    try:
+        p = await asyncio.create_subprocess_exec(
+            *cmd,
+            stdout=asyncio.subprocess.PIPE,
+            stderr=asyncio.subprocess.PIPE,
+        )
+    except FileNotFoundError:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Executable not found: {cmd[0]!r}. Is {req.backend} installed?",
+        ) from None
+    except Exception as exc:
+        raise HTTPException(status_code=500, detail=str(exc)) from exc
+
+    proc._proc = p
+    proc.pid = p.pid
+    proc.status = "running"
+    proc.log_tail.append(f"[launcher] started PID {p.pid}")
+
+    _registry(request).add(proc)
+    _task = asyncio.create_task(_tail_logs(proc))
+    _background_tasks.add(_task)
+    _task.add_done_callback(_background_tasks.discard)
+
+    return {"id": proc_id, "pid": p.pid, "command": cmd}
+
+
+@router.post("/api/launcher/stop/{proc_id}")
+async def api_stop(proc_id: str, request: Request) -> dict[str, Any]:
+    """Terminate a running process (SIGTERM, then SIGKILL after 5s)."""
+    try:
+        proc = _registry(request).get(proc_id)
+    except KeyError:
+        raise HTTPException(
+            status_code=404, detail=f"Process {proc_id!r} not found.") from None
+
+    if proc._proc and proc.status == "running":
+        proc._proc.terminate()
+        proc.log_tail.append("[launcher] SIGTERM sent")
+        try:
+            await asyncio.wait_for(proc._proc.wait(), timeout=5.0)
+        except TimeoutError:
+            proc._proc.kill()
+            proc.log_tail.append("[launcher] SIGKILL sent (timeout)")
+        proc.status = "stopped"
+        proc.pid = None
+
+    return {"id": proc_id, "status": proc.status}
+
+
+@router.delete("/api/launcher/processes/{proc_id}")
+async def api_delete(proc_id: str, request: Request) -> dict[str, Any]:
+    """Remove a stopped process from the registry."""
+    reg = _registry(request)
+    try:
+        proc = reg.get(proc_id)
+    except KeyError:
+        raise HTTPException(
+            status_code=404, detail=f"Process {proc_id!r} not found.") from None
+    if proc.status == "running":
+        raise HTTPException(status_code=400, detail="Stop the process before deleting.")
+    reg.remove(proc_id)
+    return {"deleted": proc_id}
+
+
+@router.get("/api/launcher/logs/{proc_id}")
+async def api_logs(proc_id: str, request: Request, n: int = 100) -> dict[str, Any]:
+    """Return the last N log lines for a process."""
+    try:
+        proc = _registry(request).get(proc_id)
+    except KeyError:
+        raise HTTPException(
+            status_code=404, detail=f"Process {proc_id!r} not found.") from None
+    tail = list(proc.log_tail)
+    return {"id": proc_id, "logs": tail[-n:], "total": len(tail)}
+
+
+@router.get("/api/launcher/suggest")
+async def api_suggest(model_path: str = "") -> dict[str, Any]:
+    """Suggest launch flags for the given model based on detected hardware.
+
+    クライアントの「推奨値」ボタンから呼ばれる。値はあくまで目安。
+    """
+    hw = await asyncio.to_thread(_detect_hardware)
+    size_gb = 0.0
+    if model_path:
+        size_gb = await asyncio.to_thread(_model_size_gb, model_path)
+    return {
+        "extra_args": _suggest_launch_flags(size_gb, hw),
+        "hardware": hw,
+        "size_gb": round(size_gb, 2),
+    }
+
+
+# ---------------------------------------------------------------------------
+# HTML UI
+# ---------------------------------------------------------------------------
+
+_LAUNCHER_HTML = r"""<!doctype html>
+<html lang="ja">
+<head>
+  <meta charset="utf-8" />
+  <meta name="viewport" content="width=device-width, initial-scale=1" />
+  <title>CodeRouter Launcher</title>
+  <script src="https://cdn.tailwindcss.com"></script>
+  <style>
+    .dot { width:.5rem;height:.5rem;border-radius:9999px;display:inline-block; }
+    .tabnum { font-variant-numeric:tabular-nums; }
+    .log-box { font-family:monospace;font-size:.75rem;line-height:1.4;
+               overflow-y:auto;max-height:14rem;white-space:pre-wrap;word-break:break-all; }
+    .model-row:hover { background:rgba(255,255,255,.04);cursor:pointer; }
+    .model-row.selected { background:rgba(99,102,241,.15);border-left:2px solid #6366f1; }
+    input, select, textarea {
+      background:#1e293b;border:1px solid #334155;color:#f1f5f9;
+      border-radius:.375rem;padding:.35rem .6rem;width:100%;font-size:.875rem;
+      outline:none;
+    }
+    input:focus, select:focus, textarea:focus { border-color:#6366f1; }
+    .btn-primary {
+      background:#6366f1;color:#fff;padding:.4rem 1rem;border-radius:.375rem;
+      font-size:.875rem;font-weight:600;cursor:pointer;transition:background .15s;
+    }
+    .btn-primary:hover { background:#4f46e5; }
+    .btn-primary:disabled { background:#475569;cursor:not-allowed; }
+    .btn-sm {
+      padding:.25rem .6rem;border-radius:.25rem;font-size:.75rem;
+      cursor:pointer;font-weight:500;transition:background .15s;
+    }
+    .btn-red { background:#7f1d1d;color:#fca5a5; }
+    .btn-red:hover { background:#991b1b; }
+    .btn-slate { background:#334155;color:#94a3b8; }
+    .btn-slate:hover { background:#475569; }
+    .btn-indigo { background:#312e81;color:#a5b4fc; }
+    .btn-indigo:hover { background:#3730a3; }
+  </style>
+</head>
+<body class="bg-slate-950 text-slate-100 min-h-screen font-sans">
+
+<!-- Header -->
+<header class="border-b border-slate-800 px-6 py-3">
+  <div class="max-w-7xl mx-auto flex items-center gap-x-6 text-sm">
+    <span class="text-lg font-semibold tracking-tight">CodeRouter</span>
+    <a href="/dashboard" class="text-slate-400 hover:text-slate-200 transition-colors">Dashboard</a>
+    <span class="text-slate-100 font-medium border-b border-indigo-400 pb-0.5">Launcher</span>
+    <span id="status-msg" class="ml-auto text-xs text-slate-500"></span>
+  </div>
+</header>
+
+<main class="max-w-7xl mx-auto p-4 md:p-6 space-y-4">
+
+  <!-- Row 1: Models + Launch form -->
+  <div class="grid grid-cols-1 md:grid-cols-2 gap-4">
+
+    <!-- Models panel -->
+    <section class="bg-slate-900/60 border border-slate-800 rounded-lg p-4 flex flex-col gap-3">
+      <div class="flex items-center justify-between">
+        <div class="flex items-baseline gap-2">
+          <h2 class="text-sm font-semibold uppercase tracking-wider text-slate-400">Models</h2>
+          <span id="hw-info" class="text-xs text-slate-500"></span>
+        </div>
+        <button onclick="fetchModels()" class="btn-sm btn-slate">↻ スキャン</button>
+      </div>
+      <div id="model-dirs" class="text-xs text-slate-500 space-y-0.5"></div>
+      <div id="model-list" class="divide-y divide-slate-800 text-sm flex-1 overflow-y-auto max-h-64">
+        <div class="py-2 text-slate-500 text-xs">スキャン中…</div>
+      </div>
+    </section>
+
+    <!-- Launch form -->
+    <section class="bg-slate-900/60 border border-slate-800 rounded-lg p-4 flex flex-col gap-3">
+      <h2 class="text-sm font-semibold uppercase tracking-wider text-slate-400">Launch</h2>
+
+      <div class="grid grid-cols-2 gap-2">
+        <div>
+          <label class="block text-xs text-slate-400 mb-1">名前</label>
+          <input id="f-name" type="text" placeholder="my-qwen" />
+        </div>
+        <div>
+          <label class="block text-xs text-slate-400 mb-1">ポート</label>
+          <input id="f-port" type="number" value="8080" min="1024" max="65535" />
+        </div>
+      </div>
+
+      <div>
+        <label class="block text-xs text-slate-400 mb-1">バックエンド</label>
+        <select id="f-backend" onchange="onBackendChange()">
+          <option value="llama.cpp">llama.cpp</option>
+          <option value="vllm">vllm</option>
+        </select>
+        <div id="binary-hint" class="mt-1 text-xs text-slate-500 min-h-[1.2rem]"></div>
+      </div>
+
+      <div>
+        <label class="block text-xs text-slate-400 mb-1">モデルパス</label>
+        <input id="f-model" type="text" placeholder="← モデル一覧から選択 or 直接入力" />
+      </div>
+
+      <div>
+        <label class="block text-xs text-slate-400 mb-1">オプションプロファイル</label>
+        <select id="f-profile" onchange="onProfileChange()">
+          <option value="">-- なし --</option>
+        </select>
+        <div id="profile-args" class="mt-1 text-xs font-mono text-slate-400 bg-slate-800/50 rounded p-2 hidden"></div>
+      </div>
+
+      <div>
+        <div class="flex items-center justify-between mb-1">
+          <label class="block text-xs text-slate-400">追加オプション(自由入力)</label>
+          <button onclick="suggestOptions()" class="btn-sm btn-slate">⚙ 推奨値</button>
+        </div>
+        <input id="f-extra" type="text" placeholder="-ngl 99 --threads 8" />
+      </div>
+
+      <button id="btn-launch" onclick="launchProcess()" class="btn-primary w-full mt-1">
+        ▶ 起動
+      </button>
+      <div id="launch-err" class="text-xs text-red-400 hidden"></div>
+    </section>
+  </div>
+
+  <!-- Row 2: Running processes -->
+  <section class="bg-slate-900/60 border border-slate-800 rounded-lg p-4">
+    <h2 class="text-sm font-semibold uppercase tracking-wider text-slate-400 mb-3">Processes</h2>
+    <div class="overflow-x-auto">
+      <table class="w-full text-sm tabnum">
+        <thead class="text-slate-500 text-left">
+          <tr>
+            <th class="pb-2 font-medium">NAME</th>
+            <th class="pb-2 font-medium">BACKEND</th>
+            <th class="pb-2 font-medium">MODEL</th>
+            <th class="pb-2 font-medium text-right">PORT</th>
+            <th class="pb-2 font-medium text-right">PID</th>
+            <th class="pb-2 font-medium">STATUS</th>
+            <th class="pb-2 font-medium text-right">ACTIONS</th>
+          </tr>
+        </thead>
+        <tbody id="proc-table" class="divide-y divide-slate-800">
+          <tr><td colspan="7" class="py-3 text-slate-500 text-xs">プロセスなし</td></tr>
+        </tbody>
+      </table>
+    </div>
+  </section>
+
+  <!-- Row 3: Log viewer (hidden until a process is selected) -->
+  <section id="log-panel" class="bg-slate-900/60 border border-slate-800 rounded-lg p-4 hidden">
+    <div class="flex items-center justify-between mb-2">
+      <h2 class="text-sm font-semibold uppercase tracking-wider text-slate-400">
+        Log: <span id="log-title" class="text-slate-200 normal-case">—</span>
+      </h2>
+      <div class="flex gap-2">
+        <button onclick="refreshLogs()" class="btn-sm btn-slate">↻ 更新</button>
+        <button onclick="closeLog()" class="btn-sm btn-slate">✕ 閉じる</button>
+      </div>
+    </div>
+    <div id="log-box" class="log-box bg-slate-950 rounded p-3 text-slate-300"></div>
+  </section>
+
+</main>
+
+<script>
+(() => {
+  "use strict";
+
+  const POLL_MS = 3000;
+  let allProfiles = {};      // backend → [{name, args}]
+  const _modelCache = {};    // index → {path, name, dir, size_gb}
+  let selectedLogId = null;
+  let logAutoScroll = true;
+  let _lastAutoName = "";    // selectModel が自動入力した名前
+
+  // ── Helpers ──────────────────────────────────────────────────────────────
+
+  const esc = (s) => String(s ?? "").replace(/[&<>"']/g, c =>
+    ({"&":"&amp;","<":"&lt;",">":"&gt;",'"':"&quot;","'":"&#39;"}[c])
+  );
+
+  const statusMsg = (msg, ok = true) => {
+    const el = document.getElementById("status-msg");
+    el.textContent = msg;
+    el.className = "ml-auto text-xs " + (ok ? "text-slate-500" : "text-red-400");
+    if (ok) setTimeout(() => { if (el.textContent === msg) el.textContent = ""; }, 3000);
+  };
+
+  const showLaunchErr = (msg) => {
+    const el = document.getElementById("launch-err");
+    if (msg) { el.textContent = msg; el.classList.remove("hidden"); }
+    else { el.textContent = ""; el.classList.add("hidden"); }
+  };
+
+  const statusDot = (status) => {
+    const map = {running:"bg-green-500", starting:"bg-yellow-500",
+                 stopped:"bg-slate-500", error:"bg-red-500"};
+    return `<span class="dot ${map[status] || "bg-slate-500"} mr-1.5"></span>${esc(status)}`;
+  };
+
+  // ── Models ───────────────────────────────────────────────────────────────
+
+  window.fetchModels = async () => {
+    statusMsg("モデルスキャン中…");
+    try {
+      const r = await fetch("/api/launcher/models");
+      const d = await r.json();
+      renderModelDirs(d.model_dirs || []);
+      renderHwInfo(d.hardware);
+      renderModels(d.models || []);
+      statusMsg(`モデル ${d.models.length} 件`);
+    } catch (e) {
+      statusMsg("モデルスキャン失敗: " + e.message, false);
+    }
+  };
+
+  const renderHwInfo = (hw) => {
+    const el = document.getElementById("hw-info");
+    if (!el) return;
+    if (!hw) { el.textContent = ""; return; }
+    const gpu = {metal: "Metal", cuda: "CUDA", cpu: "CPU"}[hw.gpu] || "CPU";
+    let s = `${gpu} · RAM ${hw.ram_gb}GB`;
+    if (hw.gpu === "cuda" && hw.vram_gb) s += ` · VRAM ${hw.vram_gb}GB`;
+    el.textContent = s;
+  };
+
+  const renderModelDirs = (dirs) => {
+    const el = document.getElementById("model-dirs");
+    el.innerHTML = dirs.length
+      ? dirs.map(d => `<div class="truncate">📂 ${esc(d)}</div>`).join("")
+      : '<div class="text-slate-600">model_dirs 未設定 (providers.yaml)</div>';
+  };
+
+  const recBadge = (rec) => {
+    if (!rec || !rec.label) return "";
+    if (rec.level === "ok")
+      return `<span class="text-xs shrink-0" style="color:#22c55e">✓ ${esc(rec.label)}</span>`;
+    if (rec.level === "warn")
+      return `<span class="text-xs shrink-0" style="color:#eab308">⚠ ${esc(rec.label)}</span>`;
+    return "";
+  };
+
+  const renderModels = (models) => {
+    const el = document.getElementById("model-list");
+    if (!models.length) {
+      el.innerHTML = '<div class="py-2 text-slate-500 text-xs">モデルが見つかりません</div>';
+      return;
+    }
+    el.innerHTML = models.map((m, i) => {
+      _modelCache[i] = m;
+      return `
+      <div class="model-row px-1 py-2" onclick="selectModel(${i})">
+        <div class="flex justify-between items-baseline gap-2">
+          <span class="truncate">${esc(m.name)}</span>
+          <span class="flex items-baseline gap-2 shrink-0">
+            ${recBadge(m.recommendation)}
+            <span class="text-slate-400 tabnum">${m.size_gb} GB</span>
+          </span>
+        </div>
+        <div class="text-slate-500 text-xs truncate">${esc(m.dir)}</div>
+      </div>`;
+    }).join("");
+  };
+
+  window.suggestOptions = async () => {
+    const model = document.getElementById("f-model").value.trim();
+    if (!model) { showLaunchErr("先にモデルを選択してください"); return; }
+    try {
+      const r = await fetch("/api/launcher/suggest?model_path="
+                            + encodeURIComponent(model));
+      const d = await r.json();
+      if (!r.ok) { showLaunchErr(d.detail || "推奨値の取得に失敗"); return; }
+      document.getElementById("f-extra").value = d.extra_args;
+      showLaunchErr("");
+      statusMsg("推奨値を設定(目安): " + d.extra_args);
+    } catch (e) {
+      showLaunchErr(e.message);
+    }
+  };
+
+  window.selectModel = (idx) => {
+    const m = _modelCache[idx];
+    if (!m) return;
+    document.getElementById("f-model").value = m.path;
+    // 名前が空 or 前回自動入力した値のまま → 選択モデル名で更新(手入力は保護)
+    const nameEl = document.getElementById("f-name");
+    if (!nameEl.value || nameEl.value === _lastAutoName) {
+      _lastAutoName = m.name.replace(/\.[^.]+$/, "").slice(0, 30);
+      nameEl.value = _lastAutoName;
+    }
+    document.querySelectorAll(".model-row").forEach((r, i) => {
+      r.classList.toggle("selected", i === idx);
+    });
+  };
+
+  // ── Backends (binary paths) ───────────────────────────────────────────────
+
+  let allBackends = {};
+
+  const fetchBackends = async () => {
+    try {
+      const r = await fetch("/api/launcher/backends");
+      if (!r.ok) throw new Error(`HTTP ${r.status}`);
+      const d = await r.json();
+      allBackends = d.backends || {};
+    } catch (e) {
+      console.error("[Launcher] fetchBackends failed:", e);
+    }
+    renderBinaryHint();  // always call outside try-catch so errors surface
+  };
+
+  const renderBinaryHint = () => {
+    const backend = document.getElementById("f-backend").value;
+    const hint = document.getElementById("binary-hint");
+    const btn = document.getElementById("btn-launch");
+    const info = allBackends[backend];
+    if (!info) {
+      hint.innerHTML = '<span class="text-slate-600 text-xs">バイナリ確認中…</span>';
+      return;
+    }
+    const dotColor = info.found ? "#22c55e" : "#ef4444";   // green-500 / red-500
+    const dot = `<svg style="display:inline;vertical-align:middle;margin-right:5px;flex-shrink:0" width="8" height="8" viewBox="0 0 8 8"><circle cx="4" cy="4" r="4" fill="${dotColor}"/></svg>`;
+    const label = info.is_custom ? "カスタム設定" : "PATH";
+    const statusText = info.found ? "利用可" : "見つかりません";
+    const pathColor = info.found
+      ? (info.is_custom ? "#818cf8" : "#4ade80")  // indigo-400 / green-400
+      : "#f87171";  // red-400
+    hint.innerHTML = dot
+      + `<span style="font-family:monospace;color:${pathColor};overflow:hidden;text-overflow:ellipsis;white-space:nowrap">${esc(info.resolved)}</span>`
+      + `<span style="color:#64748b;margin-left:6px;white-space:nowrap;flex-shrink:0">(${label} — ${statusText})</span>`;
+    hint.style.cssText = "display:flex;align-items:center;gap:0;overflow:hidden";
+    // Enable/disable launch button based on binary availability
+    if (!info.found) {
+      btn.disabled = true;
+      showLaunchErr(`⚠ "${esc(info.resolved)}" が見つかりません。llama.cpp をインストールするか、providers.yaml の launcher.backends.llama\\.cpp.binary にフルパスを設定してください。`);
+    } else {
+      btn.disabled = false;
+      // Clear error only if it was a binary-not-found error
+      const errEl = document.getElementById("launch-err");
+      if (errEl.textContent.startsWith("⚠")) showLaunchErr("");
+    }
+  };
+
+  // ── Option profiles ──────────────────────────────────────────────────────
+
+  const fetchProfiles = async () => {
+    try {
+      const r = await fetch("/api/launcher/option-profiles");
+      const d = await r.json();
+      allProfiles = d.profiles || {};
+      populateProfileSelect();
+      // Show hint if profiles are empty (misconfigured YAML)
+      if (d._note) console.warn("[Launcher] option-profiles:", d._note);
+    } catch (e) {
+      console.error("[Launcher] fetchProfiles error:", e);
+    }
+  };
+
+  const populateProfileSelect = () => {
+    const backend = document.getElementById("f-backend").value;
+    const sel = document.getElementById("f-profile");
+    const profiles = allProfiles[backend] || [];
+    const hint = profiles.length === 0
+      ? '<option value="" disabled style="color:#64748b">providers.yaml に option_profiles を追加すると選べます</option>'
+      : '';
+    sel.innerHTML = '<option value="">-- なし --</option>' + hint +
+      profiles.map((p, i) => `<option value="${i}">${esc(p.name)}</option>`).join("");
+    renderProfileArgs();
+  };
+
+  window.onBackendChange = () => {
+    populateProfileSelect();
+    renderBinaryHint();
+  };
+
+  // renderProfileArgs は下で const 宣言されるため、宣言前参照(TDZ)を避けて
+  // 呼び出し時に解決されるラッパーにする。
+  window.onProfileChange = () => renderProfileArgs();
+
+  const renderProfileArgs = () => {
+    const backend = document.getElementById("f-backend").value;
+    const idx = document.getElementById("f-profile").value;
+    const box = document.getElementById("profile-args");
+    if (idx === "") { box.classList.add("hidden"); box.textContent = ""; return; }
+    const profiles = allProfiles[backend] || [];
+    const p = profiles[parseInt(idx)];
+    if (!p) { box.classList.add("hidden"); return; }
+    const lines = Object.entries(p.args).map(([k, v]) =>
+      typeof v === "boolean" ? (v ? k : `# ${k} (disabled)`) : `${k} ${v}`
+    );
+    box.textContent = lines.join("  ");
+    box.classList.remove("hidden");
+  };
+
+  const selectedProfileArgs = () => {
+    const backend = document.getElementById("f-backend").value;
+    const idx = document.getElementById("f-profile").value;
+    if (idx === "") return {};
+    const profiles = allProfiles[backend] || [];
+    const p = profiles[parseInt(idx)];
+    return p ? p.args : {};
+  };
+
+  // ── Launch ───────────────────────────────────────────────────────────────
+
+  window.launchProcess = async () => {
+    showLaunchErr("");
+    const name = document.getElementById("f-name").value.trim();
+    const port = parseInt(document.getElementById("f-port").value);
+    const backend = document.getElementById("f-backend").value;
+    const model = document.getElementById("f-model").value.trim();
+    const extra = document.getElementById("f-extra").value.trim();
+
+    if (!name) { showLaunchErr("名前を入力してください"); return; }
+    if (!model) { showLaunchErr("モデルパスを入力してください"); return; }
+    if (!port || port < 1024 || port > 65535) { showLaunchErr("ポートは 1024-65535"); return; }
+
+    const btn = document.getElementById("btn-launch");
+    btn.disabled = true;
+    btn.textContent = "起動中…";
+
+    try {
+      const res = await fetch("/api/launcher/start", {
+        method: "POST",
+        headers: {"Content-Type": "application/json"},
+        body: JSON.stringify({name, backend, model_path: model, port,
+                              options: selectedProfileArgs(), extra_args: extra}),
+      });
+      const d = await res.json();
+      if (!res.ok) { showLaunchErr(d.detail || "起動失敗"); return; }
+      statusMsg(`起動: ${name} (PID ${d.pid})`);
+      // reset form name/port only
+      document.getElementById("f-name").value = "";
+      document.getElementById("f-port").value = String(port + 1);
+    } catch (e) {
+      showLaunchErr(e.message);
+    } finally {
+      btn.disabled = false;
+      btn.textContent = "▶ 起動";
+    }
+  };
+
+  // ── Processes ────────────────────────────────────────────────────────────
+
+  const fetchProcesses = async () => {
+    try {
+      const r = await fetch("/api/launcher/processes");
+      const d = await r.json();
+      renderProcesses(d.processes || []);
+    } catch (_) {}
+  };
+
+  const renderProcesses = (procs) => {
+    const tbody = document.getElementById("proc-table");
+    if (!procs.length) {
+      tbody.innerHTML = '<tr><td colspan="7" class="py-3 text-slate-500 text-xs">プロセスなし</td></tr>';
+      return;
+    }
+    tbody.innerHTML = procs.map(p => {
+      const modelName = p.model_path.split("/").pop();
+      const stopBtn = p.status === "running"
+        ? `<button onclick="stopProc('${p.id}')" class="btn-sm btn-red">■ 停止</button>`
+        : "";
+      const delBtn = p.status !== "running"
+        ? `<button onclick="deleteProc('${p.id}')" class="btn-sm btn-slate ml-1">✕</button>`
+        : "";
+      const logBtn = `<button onclick="openLog('${p.id}','${esc(p.name)}')" class="btn-sm btn-indigo ml-1">📋 ログ</button>`;
+      return `<tr>
+        <td class="py-2 pr-3 font-medium">${esc(p.name)}</td>
+        <td class="py-2 pr-3 text-slate-400">${esc(p.backend)}</td>
+        <td class="py-2 pr-3 text-slate-400 truncate max-w-[10rem]" title="${esc(p.model_path)}">${esc(modelName)}</td>
+        <td class="py-2 pr-3 text-right">${p.port}</td>
+        <td class="py-2 pr-3 text-right text-slate-400">${p.pid ?? "—"}</td>
+        <td class="py-2 pr-3">${statusDot(p.status)}</td>
+        <td class="py-2 text-right whitespace-nowrap">${stopBtn}${logBtn}${delBtn}</td>
+      </tr>`;
+    }).join("");
+  };
+
+  window.stopProc = async (id) => {
+    if (!confirm("プロセスを停止しますか?")) return;
+    const r = await fetch(`/api/launcher/stop/${id}`, {method:"POST"});
+    const d = await r.json();
+    statusMsg(`停止: ${d.status}`);
+    await fetchProcesses();
+    if (selectedLogId === id) await refreshLogs();
+  };
+
+  window.deleteProc = async (id) => {
+    if (!confirm("レジストリから削除しますか?")) return;
+    await fetch(`/api/launcher/processes/${id}`, {method:"DELETE"});
+    if (selectedLogId === id) closeLog();
+    await fetchProcesses();
+  };
+
+  // ── Log viewer ───────────────────────────────────────────────────────────
+
+  window.openLog = async (id, name) => {
+    selectedLogId = id;
+    document.getElementById("log-title").textContent = name;
+    document.getElementById("log-panel").classList.remove("hidden");
+    await refreshLogs();
+  };
+
+  window.refreshLogs = async () => {
+    if (!selectedLogId) return;
+    try {
+      const r = await fetch(`/api/launcher/logs/${selectedLogId}?n=200`);
+      const d = await r.json();
+      const box = document.getElementById("log-box");
+      box.textContent = d.logs.join("\n") || "(ログなし)";
+      if (logAutoScroll) box.scrollTop = box.scrollHeight;
+    } catch (e) {
+      document.getElementById("log-box").textContent = "ログ取得失敗: " + e.message;
+    }
+  };
+
+  window.closeLog = () => {
+    selectedLogId = null;
+    document.getElementById("log-panel").classList.add("hidden");
+  };
+
+  // ── Init + polling ───────────────────────────────────────────────────────
+
+  const init = async () => {
+    await Promise.all([fetchModels(), fetchProfiles(), fetchBackends(), fetchProcesses()]);
+  };
+
+  const poll = async () => {
+    await fetchProcesses();
+    if (selectedLogId) await refreshLogs();
+  };
+
+  init();
+  setInterval(poll, POLL_MS);
+})();
+</script>
+
+</body>
+</html>
+"""
+
+
+@router.get("/launcher", response_class=HTMLResponse)
+async def launcher_page() -> HTMLResponse:
+    """Serve the launcher single-page UI."""
+    return HTMLResponse(content=_LAUNCHER_HTML)