fix: raise Ollama timeout to 300s, add model pre-warming, improve health check

- OllamaBackend enforces _MIN_TIMEOUT=300s (overrides OLLAMA_TIMEOUT env var)
- warm_model() background task loads activeblue-chat into VRAM at startup
- health/detailed reports "warming" vs "ok" via Ollama ps() API
- README updated with May 2026 changes and test coverage details

agent_service/routers/health.py

@@ -63,6 +63,8 @@ async def _get_failing_systems() -> list[str]:
             await asyncio.wait_for(llm_router._ollama.ping(), timeout=5)
         except Exception:
             failing.append('ollama')
+    elif not llm_router:
+        failing.append('ollama')
 
     return failing
 
@@ -96,13 +98,25 @@ async def health_detailed():
         except Exception as exc:
             odoo_status = f'error: {exc}'
 
-    # Ollama check
+    # Ollama check — verify reachability and that the configured model is loaded
     ollama_status = 'unavailable'
     llm_router = get_llm_router()
     if llm_router and hasattr(llm_router, '_ollama'):
         try:
             await asyncio.wait_for(llm_router._ollama.ping(), timeout=5)
-            ollama_status = 'ok'
+            # Check whether the model is already warm in VRAM
+            import ollama as _ollama_pkg
+            client = _ollama_pkg.AsyncClient(host=llm_router._ollama._url)
+            try:
+                ps_resp = await asyncio.wait_for(client.ps(), timeout=5)
+                loaded = getattr(ps_resp, 'models', ps_resp) if not isinstance(ps_resp, dict) else ps_resp.get('models', [])
+                model_names = [getattr(m, 'model', None) or (m.get('model') if isinstance(m, dict) else None) for m in loaded]
+                if any(llm_router._ollama._model in (n or '') for n in model_names):
+                    ollama_status = 'ok'
+                else:
+                    ollama_status = 'warming'
+            except Exception:
+                ollama_status = 'ok'  # ps() unsupported — treat as ok if ping succeeded
         except Exception as exc:
             ollama_status = f'error: {exc}'