MultiagentPersonalAssistant

agent/models.py

@@ -15,13 +15,16 @@ llm_supervisor = init_chat_model(
 )
 
 llm_peripheral = init_chat_model(
-   model="groq:gemma2-9b-it"
+    model="groq:gemma2-9b-it",
+    max_tokens=4000
 )
 
 llm_agents = init_chat_model(
-    model="groq:qwen/qwen3-32b"
+    model="groq:qwen/qwen3-32b",
+    max_tokens=4000
 )
 
 llm_image = init_chat_model(
-    model="groq:meta-llama/llama-4-scout-17b-16e-instruct"
+    model="groq:meta-llama/llama-4-scout-17b-16e-instruct",
+    max_tokens=3000
 )

agent/prompts.py

@@ -23,22 +23,24 @@ deep_research_instructions = f"""You are a deep research expert. Follow these st
 """
 
 def supervisor_instructions(tools: list, agents: list):
-    return f"""You are the **Supervisor Agent**.  
-Your role is to interpret the user’s request and decide the best way to fulfill it by delegating tasks to available tools and agents.  
+    return f"""You are the Supervisor Agent. Your job is to interpret the user’s request and delegate tasks to the available tools and agents to complete the task.
 
-You have access to the following:  
+You have access to the following agents and tools:
+- Tools: {[tool.name for tool in tools]}
+- Agents: {[agent.name for agent in agents]}
 
-- **Tools:** {[tool.name for tool in tools]}  
-- **Agents:** {[agent.name for agent in agents]}  
+1. Understand and break down user’s request into smaller sub-tasks.
+2. Always delegate each sub-task to the best fitting agent/tool (e.g. always delegate any kind of coding or GitHub related task to ‘coder_agent’).
+3. Coordinate tools/agents efficiently, ensuring minimal redundancy.  
+4. Validate results before presenting them.  
+5. Respond in Telegram Markdown.  
+    - Use `*bold*`, `_italic_`, `[text](http://example.com)`, `` `inline code` ``, and ``` fenced code blocks ```.  
+    - Never leave an opening `*`, `_`, `` ` ``, `[`, or ``` without its closing pair.  
+    - Do not nest Markdown entities.  
+    - Escape reserved characters (`* _ [ ] ( ) \ ``) if they are not part of a valid entity.  
+    - Escaping inside entities is not allowed, so entity must be closed first and reopened again: use _snake_\__case_ for italic snake_case and *2*\**2=4* for bold 2*2=4.
 
-Guidelines:  
-1. Analyze the user’s request carefully.  
-2. Choose the most suitable tools and/or agents for the task.  
-3. Coordinate their use in the most efficient way to produce a clear, correct, and complete answer.  
-4. If multiple steps are required, break them down logically.  
-5. Always ensure the final output is directly useful to the user.  
-
-Your primary objective: **satisfy the user’s query as effectively as possible using the available resources.**
+Your top priority: satisfy the user’s query as effectively and efficiently as possible using the resources at your disposal, while formatting the final output in valid Markdown.
 """
 
 def coder_system_message(state: dict):

requirements.txt

@@ -34,6 +34,7 @@ PyGithub>=2.7.0
 langchain-openai>=0.3.28
 langchain-google-genai
 gtts>=2.5.4
+jiter
 timezonefinder>=7.0.1
 geocoder>=1.38.1
 aiosqlite>=0.21.0