Better agent model (#130)

* better agent

* messages

---------

Co-authored-by: balogh.adam@icloud.com <adambalogh@mac.mynetworksettings.com>

Author: adambalogh

agent/agent_executors.py

@@ -60,9 +60,8 @@
 # OpenRouter LLM Configuration
 ##
 
-GOOGLE_GEMINI_20_FLASH_MODEL = (
-    "gemini-2.5-flash"
-)
+GOOGLE_GEMINI_25_PRO_MODEL = "gemini-2.5-pro"
+GOOGLE_GEMINI_25_FLASH_MODEL = "gemini-2.5-flash"
 GROK_MODEL = "x-ai/grok-2-1212"  # $2/M input tokens; $10/M output tokens
 
 x402_http_client = x402HttpxClientv2(
@@ -75,8 +74,8 @@
 )
 
 # Select model based on configuration
-SUGGESTIONS_MODEL = GOOGLE_GEMINI_20_FLASH_MODEL
-REASONING_MODEL = GOOGLE_GEMINI_20_FLASH_MODEL
+SUGGESTIONS_MODEL = GOOGLE_GEMINI_25_FLASH_MODEL
+REASONING_MODEL = GOOGLE_GEMINI_25_PRO_MODEL
 
 
 def create_suggestions_model() -> BaseChatModel:
@@ -95,13 +94,13 @@ def create_suggestions_model() -> BaseChatModel:
 def create_investor_executor() -> any:
     openai_model = ChatOpenAI(
         model=REASONING_MODEL,
-        temperature=0.0,
+        temperature=0.1,
         api_key=config.DUMMY_X402_API_KEY,
         http_async_client=x402_http_client,
         stream_usage=True,
         streaming=True,
         base_url=config.LLM_SERVER_URL,
-        max_tokens=4096,
+        max_tokens=16384,
     )
 
     agent_executor = create_react_agent(
@@ -117,8 +116,8 @@ def create_analytics_executor(
 ) -> any:
     openai_model = ChatOpenAI(
         model=REASONING_MODEL,
-        temperature=0.0,
-        max_tokens=4096,
+        temperature=0.1,
+        max_tokens=16384,
         api_key=config.DUMMY_X402_API_KEY,
         http_async_client=x402_http_client,
         stream_usage=True,

server/activity_tracker.py

@@ -22,7 +22,7 @@ class ActivityStats:
 class PointsConfig:
     POINTS_PER_MESSAGE = 0
     POINTS_PER_SUCCESSFUL_INVITE = 0
-    DAILY_MESSAGE_LIMIT = 30
+    DAILY_MESSAGE_LIMIT = 20
 
 
 class ActivityTracker:

server/fastapi_server.py

@@ -10,6 +10,7 @@
 from fastapi.responses import JSONResponse, FileResponse
 from pydantic import ValidationError
 from langchain_core.runnables.config import RunnableConfig
+from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
 from datadog import initialize, statsd
 import aiohttp
 
@@ -574,9 +575,9 @@ async def handle_investor_chat_request(
 
     # Create messages for investor agent
     investor_messages = [
-        ("system", investor_system_prompt),
+        SystemMessage(content=investor_system_prompt),
         *message_history,
-        ("user", request.message.message),
+        HumanMessage(content=request.message.message),
     ]
 
     # Create config for the agent
@@ -677,7 +678,7 @@ async def run_main_agent(
         raise
 
 
-def convert_to_agent_message_history(messages: List[Message]) -> List[Tuple[str, str]]:
+def convert_to_agent_message_history(messages: List[Message]) -> List:
     # Get the last NUM_MESSAGES_TO_KEEP messages
     recent_messages = messages[-NUM_MESSAGES_TO_KEEP:]
 
@@ -692,8 +693,8 @@ def convert_to_agent_message_history(messages: List[Message]) -> List[Tuple[str,
             convert_to_agent_msg(recent_messages[-1], truncate=False)
         )
 
-    for _, message in converted_messages:
-        if not message:
+    for msg in converted_messages:
+        if not msg.content:
             logging.error(
                 f"Empty message.\nOriginal: {messages}\nConverted: {converted_messages}"
             )
@@ -721,9 +722,9 @@ async def handle_analytics_chat_request(
 
     # Create messages for analytics agent
     analytics_messages = [
-        ("system", analytics_system_prompt),
+        SystemMessage(content=analytics_system_prompt),
         *message_history,
-        ("user", request.message.message),
+        HumanMessage(content=request.message.message),
     ]
 
     # Create config for the agent

server/utils.py

@@ -1,26 +1,22 @@
-from typing import Tuple, List
+from typing import Tuple, List, Union
 import re
 import json
 
+from langchain_core.messages import HumanMessage, AIMessage, BaseMessage
 from api.api_types import Message, UserMessage, AgentMessage
 
 
 def convert_to_agent_msg(
     message: Message, truncate=False, max_length=800
-) -> Tuple[str, str]:
+) -> BaseMessage:
     if isinstance(message, UserMessage):
-        return ("user", message.message)
+        return HumanMessage(content=message.message)
     elif isinstance(message, AgentMessage):
         if truncate and len(message.message) > max_length:
             message_to_return = message.message[:max_length] + "... [truncated]"
         else:
             message_to_return = message.message
 
-        # if len(message.pools) > 0:
-        #     message_to_return += "\n"
-        #     for pool in message.pools:
-        #         message_to_return += f"```pool:{pool.id}```\n"
-
         if len(message.tokens) > 0:
             message_to_return += "\nTokens:\n"
             token_strings = []
@@ -36,7 +32,7 @@ def convert_to_agent_msg(
                 token_strings.append(json.dumps(token_dict))
             message_to_return += "\n- ".join(token_strings)
 
-        return ("assistant", message_to_return)
+        return AIMessage(content=message_to_return)
 
 
 def extract_patterns(