feat: add optional ollama unload after response, PR credit to @aestheticjmack

README.md

@@ -44,8 +44,8 @@ Configure via the environment variables:
 - `LLM_CONTEXT_SHALLOW_COUNT`: Results with headlines only (additional breadth). Default `15`.
 - `LLM_TABS`: Tab whitelist, comma delimiter. Default `general,science,it,news`.
 - `LLM_INTERACTIVE`: UI mode. Default is `true` (interactive: copy, regenerate, follow up). Set to `false` for simple response only mode.
-  * `LLM_OLLAMA_UNLOAD_AFTER`: If true, unload Ollama model immediately after each response (calls `/api/chat` with `keep_alive: 0`).
+- `LLM_QUESTION_MARK_REQUIRED`: Only trigger AI answers when the query contains `?`. Default `false`.
-  * `LLM_OLLAMA_UNLOAD_URL`: Override unload endpoint (default derived from `LLM_URL` host/port).
+- `LLM_OLLAMA_UNLOAD_AFTER`: Unload Ollama model after each response. Default `false`.
 
 ## How It Works
 1 user initial search 

ai_answers.py

@@ -17,11 +17,6 @@ TOKEN_EXPIRY_SEC = 3600
 STREAM_CHUNK_SIZE = 128
 STREAM_TIMEOUT_SEC = 60
 
-def _env_flag(name: str, default: bool = False) -> bool:
-    v = os.getenv(name)
-    if v is None:
-        return default
-    return v.strip().lower() in ('1', 'true', 'yes', 'on')
 def _get_streaming_connection(url: str):
     parsed = urlparse(url)
     host = parsed.hostname
@@ -482,6 +477,7 @@ class SXNGPlugin(Plugin):
 
     def _load_config(self):
         self.interactive = os.getenv('LLM_INTERACTIVE', 'true').lower().strip() in ('true', '1', 'yes', 'on')
+        self.question_mark_required = os.getenv('LLM_QUESTION_MARK_REQUIRED', 'false').lower().strip() in ('true', '1', 'yes', 'on')
         raw_provider = os.getenv('LLM_PROVIDER', '').lower().strip()
         
         raw_url = os.getenv('LLM_URL', '').strip()
@@ -553,16 +549,11 @@ class SXNGPlugin(Plugin):
         self.endpoint_url = raw_url
         
 
-        # Ollama: optional "unload after response" behavior (plugin-specific).
+        # Ollama: optional "unload after response" (frees VRAM between queries).
-        # Enable with:
+        # Enable with: LLM_OLLAMA_UNLOAD_AFTER=true
-        #   LLM_OLLAMA_UNLOAD_AFTER=1
+        self.ollama_unload_after = os.getenv('LLM_OLLAMA_UNLOAD_AFTER', 'false').lower().strip() in ('true', '1', 'yes', 'on')
-        # Optional override:
+        self.ollama_unload_url = ''
-        #   LLM_OLLAMA_UNLOAD_URL=http://<host>:11434/api/chat
+        if self.provider == 'ollama' and self.ollama_unload_after:
-        self.ollama_unload_after = (
-            _env_flag('LLM_OLLAMA_UNLOAD_AFTER', False) or _env_flag('OLLAMA_UNLOAD_AFTER', False)
-        )
-        self.ollama_unload_url = (os.getenv('LLM_OLLAMA_UNLOAD_URL') or os.getenv('OLLAMA_UNLOAD_URL') or '').strip()
-        if self.provider == 'ollama' and self.ollama_unload_after and not self.ollama_unload_url:
             try:
                 p = urlparse(self.endpoint_url)
                 scheme = p.scheme or 'http'
@@ -980,6 +971,12 @@ class SXNGPlugin(Plugin):
             if request and hasattr(request, 'headers') and request.headers.get('X-AI-Auxiliary'):
                 return results
 
+            if request and request.form.get('format', 'html') != 'html':
+                return results
+
+            if self.question_mark_required and '?' not in search.search_query.query:
+                return results
+
             current_tabs = set(search.search_query.categories)
             if not current_tabs: current_tabs = {'general'}