diff --git a/config.cpu.yaml b/config.cpu.yaml
index 77bf02d5..92614584 100644
--- a/config.cpu.yaml
+++ b/config.cpu.yaml
@@ -46,13 +46,13 @@ llm:
     n_batch: 512
     n_ctx: 8192
     max_tokens: 4096
-    template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to excersice source critisicm as the documents are returned by a search provider that can return unrelated documents.\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT!\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly. Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge. \nQuestion: {question} Let's think this step-by-step. \n<|im_end|>\n<|im_start|> assistant\n"
+    template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to exercise source criticism as the documents are returned by a search provider that can return unrelated documents.\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly. Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge.\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT\n\nSTART OF QUESTION:\n----------\n{question}\n----------\nEND OF QUESTION\n\nLet's think this step-by-step. Answer the question in the same language as the question: \n<|im_end|>\n<|im_start|> assistant\n"
     no_ctx_template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant.<|im_end|>\n<|im_start|> user\n{question}<|im_end|>\n<|im_start|> assistant\n"
     end_separator: "<|im_end|>"
 
   ctransformer:
     model: dolphin-2.2.1-mistral-7b.Q5_K_M.gguf
-    template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to excersice source critisicm as the documents are returned by a search provider that can return unrelated documents.\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT!\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly.  Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge. \nQuestion: {question} Let's think this step-by-step. \n<|im_end|>\n<|im_start|> assistant\n"
+    template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to exercise source criticism as the documents are returned by a search provider that can return unrelated documents.\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly. Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge.\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT\n\nSTART OF QUESTION:\n----------\n{question}\n----------\nEND OF QUESTION\n\nLet's think this step-by-step. Answer the question in the same language as the question: \n<|im_end|>\n<|im_start|> assistant\n"
     no_ctx_template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant.<|im_end|>\n<|im_start|> user\n{question}<|im_end|>\n<|im_start|> assistant\n"
     end_separator: "<|im_end|>"
     config:
diff --git a/config.gpu.yaml b/config.gpu.yaml
index 4232f698..6265cbe2 100644
--- a/config.gpu.yaml
+++ b/config.gpu.yaml
@@ -21,7 +21,8 @@ vectordb:
     # chroma_server_api_default_path:
 
   weaviate:
-    # auth_client_secret:
+    # auth_client_secret:emplate: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to exercise source criticism as the documents are returned by a search provider that can return unrelated documents.\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly. Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge.\n\nSTART OF QUESTION:\n----------\n{question}\n----------\nEND OF QUESTION\n\n\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT\n\nLet's think this step-by-step. Answer the question: \n<|im_end|>\n<|im_start|> assistant\n"
+
     # url: http://localhost:8080
 
 embedding:
@@ -47,7 +48,7 @@ llm:
     n_batch: 512
     n_ctx: 8192
     max_tokens: 4096
-    template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to excersice source critisicm as the documents are returned by a search provider that can return unrelated documents.\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT!\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly.  Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge. \nQuestion: {question} Let's think this step-by-step. \n<|im_end|>\n<|im_start|> assistant\n"
+    template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to exercise source criticism as the documents are returned by a search provider that can return unrelated documents.\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly. Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge.\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT\n\nSTART OF QUESTION:\n----------\n{question}\n----------\nEND OF QUESTION\n\nLet's think this step-by-step. Answer the question in the same language as the question: \n<|im_end|>\n<|im_start|> assistant\n"
     no_ctx_template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant.<|im_end|>\n<|im_start|> user\n{question}<|im_end|>\n<|im_start|> assistant\n"
     end_separator: "<|im_end|>"
     n_gpu_layers: -1
@@ -56,7 +57,7 @@ llm:
 
   ctransformer:
     model: dolphin-2.2.1-mistral-7b.Q5_K_M.gguf
-    template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to excersice source critisicm as the documents are returned by a search provider that can return unrelated documents.\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT!\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly.  Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge. \nQuestion: {question} Let's think this step-by-step. \n<|im_end|>\n<|im_start|> assistant\n"
+    template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user. <|im_end|>\n<|im_start|> user\nUse the following documents as context to answer the question at the end. REMEMBER to exercise source criticism as the documents are returned by a search provider that can return unrelated documents.\n\nIf you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly. Detect the language of the question and make sure to use the same language that was used in the question to answer the question. Don't mention which language was used, but just answer the question directly in the same langauge.\n\nSTART OF CONTEXT: \n{context} \n\nEND OF CONTEXT\n\nSTART OF QUESTION:\n----------\n{question}\n----------\nEND OF QUESTION\n\nLet's think this step-by-step. Answer the question in the same language as the question: \n<|im_end|>\n<|im_start|> assistant\n"
     no_ctx_template: "<|im_start|> system \nYou're an AI assistant named Nextcloud Assistant.<|im_end|>\n<|im_start|> user\n{question}<|im_end|>\n<|im_start|> assistant\n"
     end_separator: "<|im_end|>"
     config:
diff --git a/context_chat_backend/chain/context.py b/context_chat_backend/chain/context.py
index 45d231e7..91f6de0f 100644
--- a/context_chat_backend/chain/context.py
+++ b/context_chat_backend/chain/context.py
@@ -29,8 +29,10 @@ def get_context_docs(
 def get_context_chunks(context_docs: list[Document]) -> list[str]:
 	context_chunks = []
 	for doc in context_docs:
+		chunk = '\n\nSTART OF DOCUMENT'
 		if title := doc.metadata.get('title'):
-			context_chunks.append(title)
-		context_chunks.append(doc.page_content)
+			chunk += '\nDocument: ' + title
+		chunk += "\n\n" + doc.page_content + "\n\nEND OF DOCUMENT"
+		context_chunks.append(chunk)
 
 	return context_chunks
diff --git a/context_chat_backend/chain/one_shot.py b/context_chat_backend/chain/one_shot.py
index 3875816b..141f07e6 100644
--- a/context_chat_backend/chain/one_shot.py
+++ b/context_chat_backend/chain/one_shot.py
@@ -10,10 +10,29 @@
 from .query_proc import get_pruned_query
 from .types import ContextException, LLMOutput, ScopeType
 
-_LLM_TEMPLATE = '''Answer based only on this context and do not add any imaginative details. Make sure to use the same language as the question in your answer.
+_LLM_TEMPLATE = '''
+You're an AI assistant named Nextcloud Assistant, good at finding relevant context from documents to answer questions provided by the user.
+Use the following documents as context to answer the question at the end. REMEMBER to exercise source criticism as the documents are returned by a search provider that can return unrelated documents.
+If you don't know the answer or are unsure, just say that you don't know, don't try to make up an answer. Don't mention the context in your answer but rather just answer the question directly. Detect the language of the question and make sure to use the same language that was used in the question to answer the question.
+Don't mention which language was used, but just answer the question directly in the same langauge.
+
+CONTEXT:
+-----------------
+
 {context}
 
+-----------------
+END OF CONTEXT
+
+QUESTION:
+-----------------
+
 {question}
+
+-----------------
+END OF QUESTION
+
+Let's think this step-by-step. Answer the question in the same language as the question:
 ''' # noqa: E501