diff --git a/pyproject.toml b/pyproject.toml
index 4b651dad..edbbb655 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -66,7 +66,7 @@ dependencies = [
     "pymupdf >= 1.23.5",
     "django == 5.0.7",
     "authlib == 1.2.1",
-    "llama-cpp-python == 0.2.82",
+    "llama-cpp-python == 0.2.88",
     "itsdangerous == 2.1.2",
     "httpx == 0.25.0",
     "pgvector == 0.2.4",
diff --git a/src/khoj/processor/conversation/offline/chat_model.py b/src/khoj/processor/conversation/offline/chat_model.py
index ec4c7367..1251dcec 100644
--- a/src/khoj/processor/conversation/offline/chat_model.py
+++ b/src/khoj/processor/conversation/offline/chat_model.py
@@ -103,6 +103,9 @@ def extract_questions_offline(
             .replace("']", '"]')
             .replace("', '", '", "')
         )
+        # Remove any markdown json codeblock formatting if present (useful for gemma-2)
+        if response.startswith("```json"):
+            response = response[7:-3]
         questions: List[str] = json.loads(questions_str)
         questions = filter_questions(questions)
     except:
diff --git a/src/khoj/processor/conversation/prompts.py b/src/khoj/processor/conversation/prompts.py
index 6a8db9db..ffd7d094 100644
--- a/src/khoj/processor/conversation/prompts.py
+++ b/src/khoj/processor/conversation/prompts.py
@@ -587,7 +587,7 @@ You are Khoj, an advanced google search assistant. You are tasked with construct
 - Official, up-to-date information about you, Khoj, is available at site:khoj.dev, github or pypi.
 
 What Google searches, if any, will you need to perform to answer the user's question?
-Provide search queries as a list of strings in a JSON object. Do not wrap the json in a codeblock.
+Provide search queries as a list of strings in a JSON object.
 Current Date: {current_date}
 User's Location: {location}
 {username}
diff --git a/src/khoj/routers/helpers.py b/src/khoj/routers/helpers.py
index 4e4f5a56..4da60717 100644
--- a/src/khoj/routers/helpers.py
+++ b/src/khoj/routers/helpers.py
@@ -279,6 +279,9 @@ async def aget_relevant_information_sources(query: str, conversation_history: di
 
     try:
         response = response.strip()
+        # Remove any markdown json codeblock formatting if present (useful for gemma-2)
+        if response.startswith("```json"):
+            response = response[7:-3]
         response = json.loads(response)
         response = [q.strip() for q in response["source"] if q.strip()]
         if not isinstance(response, list) or not response or len(response) == 0:
@@ -401,6 +404,9 @@ async def generate_online_subqueries(
     # Validate that the response is a non-empty, JSON-serializable list
     try:
         response = response.strip()
+        # Remove any markdown json codeblock formatting if present (useful for gemma-2)
+        if response.startswith("```json") and response.endswith("```"):
+            response = response[7:-3]
         response = json.loads(response)
         response = [q.strip() for q in response["queries"] if q.strip()]
         if not isinstance(response, list) or not response or len(response) == 0: