src/backend/agents/base.py

@@ -358,9 +358,8 @@ class Agent(BaseModel, ABC):
                         size=user.file_watcher.collection.count()
                     )
 
-                    entries += len(rag_metadata.documents)
                     rag_message.metadata.rag_results.append(rag_metadata)
-                    rag_message.content = f"Results from {rag.name} RAG: {len(rag_metadata.documents)} results."
+                    rag_message.content = f"Results from {rag.name} RAG: {len(chroma_results['documents'])} results."
                     yield rag_message
 
             rag_message.content = (
@@ -378,7 +377,7 @@ class Agent(BaseModel, ABC):
             return
 
     async def generate(
-        self, llm: Any, model: str, user_message: ChatMessageUser, user: Candidate, temperature=0.7
+        self, llm: Any, model: str, user_message: ChatMessageUser, user_id: str, temperature=0.7
     ) -> AsyncGenerator[ChatMessage | ChatMessageBase, None]:
         logger.info(f"{self.agent_type} - {inspect.stack()[0].function}")
 
@@ -447,7 +446,7 @@ Content: { content }
                 messages.append(
                     LLMMessage(
                         role="user",
-                        content=f"<|context|>\n{rag_context.strip()}\n</|context|>\n\nPrompt to respond to:\n{user_message.content.strip()}\n"
+                        content=f"<|context|>\n{rag_context.strip()}\n</|context|>\n\n{user_message.content.strip()}\n"
                     )
                 )
             else:

src/backend/agents/candidate_chat.py

@@ -7,20 +7,44 @@ from .base import Agent, agent_registry
 from logger import logger
 
 from .registry import agent_registry
-from models import ( ChatQuery, ChatMessage, Tunables, ChatStatusType, ChatMessageUser, Candidate)
-
+from models import ( ChatQuery, ChatMessage, Tunables, ChatStatusType)
 
 system_message = f"""
+Launched on {datetime.now().isoformat()}.
+
 When answering queries, follow these steps:
 
-- When any content from <|context|> is relevant, synthesize information from all sources to provide the most complete answer.
-- Always prioritize the most up-to-date, recent, and relevant information first.
-- If there is information in the <|context|> section to enhance the answer, incorporate it seamlessly and refer to it as 'the latest information' or 'recent data' instead of mentioning '<|context|>' (etc.) or quoting it directly.
-- Avoid phrases like 'According to the <|context|>' or similar references to the <|context|>.
+- First analyze the query to determine if real-time information from the tools might be helpful
+- Even when <|context|> or <|resume|> is provided, consider whether the tools would provide more current or comprehensive information
+- Use the provided tools whenever they would enhance your response, regardless of whether context is also available
+- When presenting weather forecasts, include relevant emojis immediately before the corresponding text. For example, for a sunny day, say \"☀️ Sunny\" or if the forecast says there will be \"rain showers, say \"🌧️ Rain showers\". Use this mapping for weather emojis: Sunny: ☀️, Cloudy: ☁️, Rainy: 🌧️, Snowy: ❄️
+- When any combination of <|context|>, <|resume|> and tool outputs are relevant, synthesize information from all sources to provide the most complete answer
+- Always prioritize the most up-to-date and relevant information, whether it comes from <|context|>, <|resume|> or tools
+- If <|context|> and tool outputs contain conflicting information, prefer the tool outputs as they likely represent more current data
+- If there is information in the <|context|> or <|resume|> sections to enhance the answer, incorporate it seamlessly and refer to it as 'the latest information' or 'recent data' instead of mentioning '<|context|>' (etc.) or quoting it directly.
+- Avoid phrases like 'According to the <|context|>' or similar references to the <|context|> or <|resume|>.
 
-Always <|context|> when possible. Be concise, and never make up information. If you do not know the answer, say so.
+CRITICAL INSTRUCTIONS FOR IMAGE GENERATION:
 
-Before answering, ensure you have spelled the candidate's name correctly.
+1. When the user requests to generate an image, inject the following into the response: <GenerateImage prompt="USER-PROMPT"/>. Do this when users request images, drawings, or visual content.
+3. MANDATORY: You must respond with EXACTLY this format: <GenerateImage prompt="{{USER-PROMPT}}"/>
+4. FORBIDDEN: DO NOT use markdown image syntax ![](url) 
+5. FORBIDDEN: DO NOT create fake URLs or file paths
+6. FORBIDDEN: DO NOT use any other image embedding format
+
+CORRECT EXAMPLE:
+User: "Draw a cat"
+Your response: "<GenerateImage prompt='Draw a cat'/>"
+
+WRONG EXAMPLES (DO NOT DO THIS):
+- ![](https://example.com/...)
+- ![Cat image](any_url)
+- <img src="...">
+
+The <GenerateImage prompt="{{USER-PROMPT}}"/> format is the ONLY way to display images in this system. 
+DO NOT make up a URL for an image or provide markdown syntax for embedding an image. Only use <GenerateImage prompt="{{USER-PROMPT}}".
+
+Always use tools, <|resume|>, and <|context|> when possible. Be concise, and never make up information. If you do not know the answer, say so.
 """
 
 class CandidateChat(Agent):
@@ -33,26 +57,31 @@ class CandidateChat(Agent):
 
     system_prompt: str = system_message
 
-    async def generate(
-        self, llm: Any, model: str, user_message: ChatMessageUser, user: Candidate, temperature=0.7
-    ):
-        self.system_prompt = f"""
-You are a helpful expert system representing a {user.first_name}'s work history to potential employers and users curious about the candidate. You want to incorporate as many facts and details about {user.first_name} as possible.
+#     async def prepare_message(self, message: Message) -> AsyncGenerator[Message, None]:
+#         logger.info(f"{self.agent_type} - {inspect.stack()[0].function}")
+#         if not self.context:
+#             raise ValueError("Context is not set for this agent.")
 
-When referencing the candidate, ALWAYS ensure correct spelling.
+#         async for message in super().prepare_message(message):
+#             if message.status != "done":
+#                 yield message
 
-The candidate's first name is: "{user.first_name}"
-The candidate's last name is: "{user.last_name}"
-
-Use that spelling instead of any spelling you may find in the <|context|>.
-
-{system_message}
-"""
-
-        async for message in super().generate(llm, model, user_message, user, temperature):
-            yield message
+#         if message.preamble:
+#             excluded = {}
+#             preamble_types = [
+#                 f"<|{p}|>" for p in message.preamble.keys() if p not in excluded
+#             ]
+#             preamble_types_AND = " and ".join(preamble_types)
+#             preamble_types_OR = " or ".join(preamble_types)
+#             message.preamble[
+#                 "rules"
+#             ] = f"""\
+# - Answer the question based on the information provided in the {preamble_types_AND} sections by incorporate it seamlessly and refer to it using natural language instead of mentioning {preamble_types_OR} or quoting it directly.
+# - If there is no information in these sections, answer based on your knowledge, or use any available tools.
+# - Avoid phrases like 'According to the {preamble_types[0]}' or similar references to the {preamble_types_OR}.
+# """
+#         message.preamble["question"] = "Respond to:"
 
 
 # Register the base agent
 agent_registry.register(CandidateChat._agent_type, CandidateChat)
-

src/backend/main.py

@@ -1994,7 +1994,7 @@ async def post_chat_session_message_stream(
                     llm=llm_manager.get_llm(),
                     model=defines.model,
                     user_message=user_message,
-                    user=current_user,
+                    user_id=current_user.id,
                 ):
                     # Store reference to the complete AI message
                     if generated_message.status == ChatStatusType.DONE: