fix(ollama): disable thinking for deepseek call sites

apps/api/src/services/alert_rule_engine.py

@@ -720,7 +720,14 @@ async def _call_ollama(prompt: str, ollama_url: str, model: str) -> str | None:
             try:
                 resp = await client.post(
                     f"{endpoint.url}/api/generate",
-                    json={"model": model, "prompt": prompt, "stream": False, "options": {"temperature": 0.1}},
+                    json={
+                        "model": model,
+                        "prompt": prompt,
+                        "stream": False,
+                        # Preserve response-body compatibility with Ollama 0.24 thinking models.
+                        "think": False,
+                        "options": {"temperature": 0.1},
+                    },
                 )
                 resp.raise_for_status()
                 return resp.json().get("response", "")

apps/api/src/services/chat_manager.py

@@ -167,6 +167,9 @@ class ChatManager:
                         json={
                             "model": MODEL,
                             "stream": False,
+                            # Ollama 0.24 separates deepseek-r1 thinking from final text.
+                            # Chat callers expect message.content to contain the answer.
+                            "think": False,
                             "messages": [
                                 {"role": "system", "content": system_prompt},
                                 {"role": "user", "content": user_message},

apps/api/src/services/decision_manager.py

@@ -678,7 +678,13 @@ async def _nemoclaw_second_opinion(incident: "Incident", primary_result: dict) -
                 try:
                     resp = await client.post(
                         f"{endpoint.url.rstrip('/')}/api/generate",
-                        json={"model": model, "prompt": prompt, "stream": False},
+                        json={
+                            "model": model,
+                            "prompt": prompt,
+                            "stream": False,
+                            # Ollama 0.24 puts deepseek-r1 output in `thinking` unless disabled.
+                            "think": False,
+                        },
                     )
                     resp.raise_for_status()
                     data = resp.json()

apps/api/src/services/log_summary_service.py

@@ -220,6 +220,9 @@ class LogSummaryService:
                             "model": SUMMARY_MODEL,
                             "prompt": prompt,
                             "stream": False,
+                            # Ollama 0.24 returns deepseek-r1 text in `thinking` by default.
+                            # Existing callers read `response`, so force final-answer mode.
+                            "think": False,
                             "options": {"temperature": 0.1, "num_predict": 200},
                         },
                     )