fix: remote model pulling + ollama path after install

Model pulling:
- Changed from CLI (ollama pull) to Ollama HTTP API (/api/pull).
  This works for both local AND remote Ollama — the pull runs on
  whatever machine Ollama is running on, not the local machine.
- Pull buttons now send the correct Ollama URL (local or remote)
- Status shows which URL the download is targeting

Remote install fix:
- Use /usr/local/bin/ollama for version check after install (PATH
  not refreshed in same SSH session)
- Added sleep after service start for stability

https://claude.ai/code/session_018kEJ1dG42MJkBYbrK68h79

Author: claude

clanker/setup/ollama.py

@@ -207,7 +207,8 @@ def install_ollama_remote(ssh_host: str) -> dict[str, Any]:
                 *ssh_args,
                 "curl -fsSL https://ollama.com/install.sh | sh && "
                 "systemctl start ollama 2>/dev/null; "
-                "ollama --version",
+                "sleep 2; "
+                "/usr/local/bin/ollama --version 2>/dev/null || ollama --version",
             ],
             capture_output=True, text=True, timeout=300,
         )
@@ -229,17 +230,28 @@ def install_ollama_remote(ssh_host: str) -> dict[str, Any]:
 def pull_model(
     model: str, base_url: str = "http://localhost:11434"
 ) -> dict[str, Any]:
-    """Pull a model via Ollama CLI (shows progress)."""
+    """Pull a model via Ollama API (works for both local and remote)."""
+    url = base_url.rstrip("/")
     try:
-        result = subprocess.run(
-            ["ollama", "pull", model],
-            capture_output=True, text=True, timeout=600,
-        )
-        if result.returncode == 0:
-            return {"ok": True, "message": f"Pulled {model}"}
-        return {"ok": False, "message": result.stderr[:500]}
-    except subprocess.TimeoutExpired:
-        return {"ok": False, "message": f"Pull of {model} timed out"}
+        # Use the Ollama API instead of CLI — works for remote too
+        with httpx.Client(timeout=600.0) as client:
+            resp = client.post(
+                f"{url}/api/pull",
+                json={"name": model, "stream": False},
+            )
+            if resp.status_code == 200:
+                return {"ok": True, "message": f"Pulled {model}"}
+            return {
+                "ok": False,
+                "message": f"Pull failed: HTTP {resp.status_code}",
+            }
+    except httpx.ReadTimeout:
+        return {"ok": False, "message": f"Pull of {model} timed out (large download)"}
+    except httpx.ConnectError:
+        return {
+            "ok": False,
+            "message": f"Cannot connect to Ollama at {url}. Is it running?",
+        }
     except Exception as exc:
         return {"ok": False, "message": str(exc)}
 

clanker/setup/static/index.html

@@ -942,12 +942,13 @@ <h2>Setup Complete!</h2>
 }
 
 async function pullModel(model) {
-  setStatus('ollama-pull-status', 'loading', 'Downloading ' + model + ' (this may take a few minutes)...');
-  const r = await api('/api/ollama/pull', {model: model});
+  const url = getOllamaUrl();
+  setStatus('ollama-pull-status', 'loading', 'Downloading ' + model + ' to ' + url + ' (this may take a few minutes)...');
+  const r = await api('/api/ollama/pull', {model: model, base_url: url});
   if (r.ok) {
     setStatus('ollama-pull-status', 'ok', 'Downloaded ' + model);
     // Refresh model list
-    const r2 = await api('/api/test/ollama', {base_url: getOllamaUrl()});
+    const r2 = await api('/api/test/ollama', {base_url: url});
     if (r2.ok && r2.models) populateSelect('ollama-model', r2.models, model);
   } else {
     setStatus('ollama-pull-status', 'err', r.message);

clanker/setup/web.py

@@ -132,7 +132,8 @@ def _handle_ollama_pull(self, body: dict[str, Any]) -> None:
         from clanker.setup.ollama import pull_model
 
         model = body.get("model", "llama3.2")
-        self._json_response(pull_model(model))
+        base_url = body.get("base_url", "http://localhost:11434")
+        self._json_response(pull_model(model, base_url=base_url))
 
     def _handle_ollama_optimize(self, body: dict[str, Any]) -> None:
         from clanker.setup.ollama import apply_systemd_env, get_optimization_advice