feat: specsmith ollama command group + GPU detection

New module: src/specsmith/ollama_cmds.py
- detect_gpu(): nvidia-smi primary, Windows WMI fallback (AMD/Intel)
- MODEL_CATALOG: 9 curated models with VRAM, size, best-for, tier, ctx
- get_installed_models/ids(): calls Ollama /api/tags
- recommend_models(vram_gb): filter by 90% VRAM budget
- suggest_for_task(task, vram_gb): scored model suggestions by task type

New CLI: specsmith ollama <command>
- list: table of installed models with size + date
- available: GPU-aware table of catalog vs installed (✓/⬇)
- gpu: detect GPU, show VRAM tier, list recommended models
- pull <model>: streaming download with progress bar, Ctrl-C cancels
- suggest <task>: ordered suggestions for coding/requirements/architecture/chat/analysis/reasoning

Co-Authored-By: Oz <oz-agent@warp.dev>

Author: tbitcs

src/specsmith/cli.py

@@ -2592,6 +2592,242 @@ def auth_check(project_dir: str) -> None:
 main.add_command(auth)
 
 
+# ---------------------------------------------------------------------------
+# Ollama — local LLM model management
+# ---------------------------------------------------------------------------
+
+
+@main.group(name="ollama")
+def ollama_group() -> None:
+    """Manage local Ollama models (list, download, GPU detection)."""
+
+
+@ollama_group.command(name="list")
+def ollama_list() -> None:
+    """List locally installed Ollama models."""
+    from specsmith.ollama_cmds import get_installed_models, is_running
+
+    if not is_running():
+        console.print("[red]✗[/red] Ollama is not running. Start it with: [bold]ollama serve[/bold]")
+        raise SystemExit(1)
+
+    models = get_installed_models()
+    if not models:
+        console.print("[yellow]No models installed.[/yellow] Run: [bold]specsmith ollama available[/bold]")
+        return
+
+    console.print(f"[bold]Installed Ollama Models[/bold] ({len(models)})\n")
+    for m in models:
+        size_gb = m.get("size", 0) / (1024**3)
+        modified = m.get("modified_at", "")[:10]
+        console.print(f"  [green]✓[/green] {m['name']:<35s} {size_gb:.1f}GB  [{modified}]")
+
+
+@ollama_group.command(name="available")
+def ollama_available() -> None:
+    """Show recommended models vs installed. GPU-aware."""
+    from specsmith.ollama_cmds import (
+        MODEL_CATALOG,
+        detect_gpu,
+        get_installed_ids,
+        gpu_tier,
+        is_running,
+    )
+
+    gpus = detect_gpu()
+    vram = max((g["vram_gb"] for g in gpus), default=0)
+    tier = gpu_tier(vram)
+    budget = vram * 0.90 if vram else 999
+
+    if gpus:
+        console.print(f"[bold]GPU:[/bold] {gpus[0]['name']} — {vram:.1f}GB VRAM ({tier})\n")
+    else:
+        console.print("[yellow]No GPU detected — CPU mode (small models only)[/yellow]\n")
+
+    running = is_running()
+    installed_ids: list[str] = []
+    if running:
+        installed_ids = get_installed_ids()
+    else:
+        console.print("[dim]Ollama not running — showing catalog only[/dim]\n")
+
+    console.print(
+        f"{'Model':<35s} {'VRAM':<8s} {'Size':<8s} {'Fits?':<7s} {'Status':<12s} Best for"
+    )
+    console.print("-" * 100)
+    for m in MODEL_CATALOG:
+        fits = m["vram_gb"] <= budget or not gpus
+        is_installed = any(m["id"] in iid or iid in m["id"] for iid in installed_ids)
+        fits_str = "[green]✓[/green]" if fits else "[dim]✓ CPU?[/dim]" if not gpus else "[red]✗ VRAM[/red]"
+        status = "[green]✓ Installed[/green]" if is_installed else "[dim]⬇ Available[/dim]"
+        best = ", ".join(m["best_for"][:2])
+        console.print(
+            f"  {m['name']:<33s} {m['vram_gb']:<8.1f} {m['size_gb']:<8.1f} "
+            f"{fits_str:<12s} {status:<20s} {best}"
+        )
+
+    console.print("\n  Run [bold]specsmith ollama pull <model-id>[/bold] to download")
+    console.print("  e.g. [bold]specsmith ollama pull qwen2.5:14b[/bold]")
+
+
+@ollama_group.command(name="gpu")
+def ollama_gpu() -> None:
+    """Show GPU information and model tier recommendations."""
+    from specsmith.ollama_cmds import detect_gpu, gpu_tier, recommend_models
+
+    gpus = detect_gpu()
+    if not gpus:
+        console.print("[yellow]No dedicated GPU detected.[/yellow]")
+        console.print("  Ollama can run on CPU but will be slow.")
+        console.print("  Recommended: llama3.2:latest (2GB RAM only)")
+        return
+
+    for gpu in gpus:
+        vram = gpu["vram_gb"]
+        tier = gpu_tier(vram)
+        console.print(f"[bold]GPU:[/bold] {gpu['name']}")
+        console.print(f"  VRAM:  {vram:.1f} GB")
+        console.print(f"  Tier:  {tier}")
+        console.print()
+
+        recs = recommend_models(vram)
+        console.print(f"  [bold]Recommended models for {vram:.1f}GB VRAM:[/bold]")
+        for m in recs:
+            console.print(
+                f"    [green]✓[/green] {m['name']:<35s} "
+                f"{m['vram_gb']:.1f}GB  —  {', '.join(m['best_for'][:2])}"
+            )
+        console.print()
+        console.print("  Run [bold]specsmith ollama pull <model-id>[/bold] to download.")
+
+
+@ollama_group.command(name="pull")
+@click.argument("model")
+def ollama_pull(model: str) -> None:
+    """Download a model from Ollama library.
+
+    MODEL: model id, e.g. qwen2.5:14b, phi4:latest
+
+    Press Ctrl-C to cancel the download.
+    """
+    import signal
+    import sys
+    import urllib.error
+
+    from specsmith.ollama_cmds import MODEL_CATALOG, is_running
+
+    if not is_running():
+        console.print("[red]✗[/red] Ollama is not running. Start it first: [bold]ollama serve[/bold]")
+        raise SystemExit(1)
+
+    # Show model info if in catalog
+    info = next((m for m in MODEL_CATALOG if m["id"] == model), None)
+    if info:
+        console.print(
+            f"[bold]Pulling[/bold] {info['name']} ({info['size_gb']:.1f} GB)\n"
+            f"  Best for: {', '.join(info['best_for'])}\n"
+            f"  Notes:    {info['notes']}\n"
+        )
+    else:
+        console.print(f"[bold]Pulling[/bold] {model} from Ollama library\n")
+
+    console.print("[dim]Press Ctrl-C to cancel[/dim]\n")
+
+    # Stream pull progress from Ollama
+    import json
+    import urllib.request
+    from specsmith.ollama_cmds import OLLAMA_API
+
+    payload = json.dumps({"name": model}).encode()
+    req = urllib.request.Request(  # noqa: S310
+        f"{OLLAMA_API}/api/pull",
+        data=payload,
+        headers={"Content-Type": "application/json"},
+    )
+
+    last_status = ""
+    try:
+        with urllib.request.urlopen(req, timeout=600) as resp:  # noqa: S310
+            for line in resp:
+                line = line.strip()
+                if not line:
+                    continue
+                try:
+                    chunk = json.loads(line)
+                    status = chunk.get("status", "")
+                    completed = chunk.get("completed", 0)
+                    total = chunk.get("total", 0)
+                    if status == "success":
+                        console.print(f"\n[bold green]✓ {model} downloaded successfully.[/bold green]")
+                        return
+                    if total and status == "pulling " + (chunk.get("digest", "")[:12] if chunk.get("digest") else ""):
+                        pct = int(completed / total * 100)
+                        bar = "█" * (pct // 5) + "░" * (20 - pct // 5)
+                        sys.stdout.write(f"\r  [{bar}] {pct:3d}%  {completed/(1024**3):.2f}/{total/(1024**3):.2f} GB")
+                        sys.stdout.flush()
+                    elif status != last_status:
+                        console.print(f"  {status}")
+                        last_status = status
+                except json.JSONDecodeError:
+                    pass
+    except KeyboardInterrupt:
+        console.print("\n[yellow]Download cancelled.[/yellow]")
+        raise SystemExit(0)
+    except urllib.error.URLError as e:
+        console.print(f"\n[red]Error: {e}[/red]")
+        raise SystemExit(1)
+
+
+@ollama_group.command(name="suggest")
+@click.argument("task", required=False)
+def ollama_suggest(task: str | None) -> None:
+    """Suggest models for a given task type.
+
+    TASK: coding | requirements | architecture | chat | analysis | reasoning
+
+    Example: specsmith ollama suggest coding
+    """
+    from specsmith.ollama_cmds import (
+        TASK_TAGS,
+        detect_gpu,
+        get_installed_ids,
+        is_running,
+        suggest_for_task,
+    )
+
+    if not task:
+        console.print("[bold]Available task types:[/bold]")
+        for key in TASK_TAGS:
+            console.print(f"  {key}")
+        console.print("\nUsage: [bold]specsmith ollama suggest <task>[/bold]")
+        return
+
+    gpus = detect_gpu()
+    vram = max((g["vram_gb"] for g in gpus), default=999)
+    suggestions = suggest_for_task(task, vram)
+
+    if not suggestions:
+        console.print(f"[yellow]No models found for task '{task}'.[/yellow]")
+        return
+
+    installed_ids: list[str] = []
+    if is_running():
+        installed_ids = get_installed_ids()
+
+    console.print(f"[bold]Model suggestions for task: {task}[/bold]\n")
+    for m in suggestions:
+        is_installed = any(m["id"] in iid or iid in m["id"] for iid in installed_ids)
+        status = "[green]✓ installed[/green]" if is_installed else f"[dim]⬇ {m['size_gb']:.1f}GB[/dim]"
+        console.print(
+            f"  {status}  [bold]{m['name']:<35s}[/bold] {m['notes']}"
+        )
+        console.print(f"    [dim]ID: {m['id']}  VRAM: {m['vram_gb']:.1f}GB  ctx: {m['ctx_k']}K[/dim]")
+    console.print("\n  Download: [bold]specsmith ollama pull <id>[/bold]")
+
+
+main.add_command(ollama_group)
+
+
 # ---------------------------------------------------------------------------
 # Workspace — multi-project management (#17)
 # ---------------------------------------------------------------------------