Skip to content

Building AI agents, atomically

Start work on a more modular dynamic way of building the agents system prompt

Kenny Vaneetvelde65cb80c1cb23
diff --git a/atomic_agents/agents/basic_chat_agent_new.py b/atomic_agents/agents/basic_chat_agent_new.py
new file mode 100644
index 0000000..1bcc8d6
--- /dev/null
+++ b/atomic_agents/agents/basic_chat_agent_new.py
@@ -0,0 +1,173 @@
+import os
+from typing import Callable, List
+from pydantic import BaseModel, Field
+from datetime import datetime
+from groq import Groq
+import instructor
+import openai
+from atomic_agents.lib.chat_memory import ChatMemory
+from rich.console import Console
+
+class BasicChatAgentInputSchema(BaseModel):
+ chat_input: str = Field(..., description='The input text for the chat agent.')
+
+class BasicChatAgentResponse(BaseModel):
+ response: str = Field(..., description='The markdown-enabled response from the chat agent.')
+
+ class Config:
+ title = 'BasicChatAgentResponse'
+ description = 'Response from the basic chat agent. The response can be in markdown format.'
+ json_schema_extra = {
+ 'title': title,
+ 'description': description
+ }
+
+class SystemPrompt(BaseModel):
+ background: List[str]
+ steps: List[str]
+ output_instructions: List[str]
+
+class DynamicInfoProvider:
+ def __init__(self, provider_func: Callable[..., str], title: str, *args, **kwargs):
+ self.provider_func = provider_func
+ self.title = title
+ self.args = args
+ self.kwargs = kwargs
+
+ def get_info(self) -> str:
+ return self.provider_func(*self.args, **self.kwargs)
+
+class SystemPromptGenerator:
+ def __init__(self, system_prompt: SystemPrompt, dynamic_info_providers: List[DynamicInfoProvider] = None):
+ self.system_prompt = system_prompt
+ self.dynamic_info_providers = dynamic_info_providers or []
+
+ def generate_prompt(self) -> str:
+ # Generate the prompt
+ system_prompt = ''
+
+ if self.system_prompt.background:
+ system_prompt += '# IDENTITY and PURPOSE\n'
+ system_prompt += f'-{'\n-'.join(self.system_prompt.background)}\n\n'
+
+ if self.system_prompt.steps:
+ system_prompt += '# Steps\n'
+ system_prompt += f'-{'\n-'.join(self.system_prompt.steps)}\n\n'
+
+ if self.system_prompt.output_instructions:
+ system_prompt += '# OUTPUT INSTRUCTIONS\n'
+ system_prompt += f'-{'\n-'.join(self.system_prompt.output_instructions)}\n\n'
+
+ if self.dynamic_info_providers:
+ system_prompt += '# EXTRA INFORMATION AND CONTEXT\n'
+ for provider in self.dynamic_info_providers:
+ system_prompt += f'## {provider.title}\n'
+ system_prompt += f'{provider.get_info()}\n\n'
+
+ return system_prompt
+
+class BasicChatAgent:
+ def __init__(self, client, model: str = 'gpt-3.5-turbo', system_prompt: SystemPrompt = None, memory: ChatMemory = None, vector_db = None, input_schema = BasicChatAgentInputSchema, output_schema = BasicChatAgentResponse, dynamic_info_providers: List[DynamicInfoProvider] = None):
+ self.input_schema = input_schema
+ self.output_schema = output_schema
+ self.client = client
+ self.model = model
+ self.memory = memory or ChatMemory()
+ self.system_prompt = system_prompt or SystemPrompt(
+ background=['You are a helpful AI assistant.'],
+ steps=[],
+ output_instructions=[]
+ )
+ self.prompt_generator = SystemPromptGenerator(self.system_prompt, dynamic_info_providers)
+
+ def get_system_prompt(self) -> str:
+ return self.prompt_generator.generate_prompt()
+
+ def get_response(self, response_model=None) -> BaseModel:
+ if response_model is None:
+ response_model = self.output_schema
+
+ messages = [{'role': 'system', 'content': self.get_system_prompt()}] + self.memory.get_history()
+ response = self.client.chat.completions.create(
+ model=self.model,
+ messages=messages,
+ response_model=response_model
+ )
+ return response
+
+ def run(self, user_input: str) -> str:
+ self.memory.add_message('user', user_input)
+ response = self.get_response(response_model=self.output_schema)
+ self.memory.add_message('assistant', response.model_dump_json())
+ return response
+
+if __name__ == '__main__':
+ console = Console()
+
+ openai_client = openai.OpenAI()
+ client = instructor.from_openai(openai_client)
+ model = 'gpt-3.5-turbo'
+ groq_client = instructor.from_groq(Groq(api_key=os.getenv('GROQ_API_KEY')))
+ groq_model = os.getenv('GROQ_CHAT_MODEL')
+
+ class MyChatAgentResponse(BaseModel):
+ observations: List[str] = Field(..., description='What the agent observes in the current state of the conversation.')
+ thought_process: List[str] = Field(..., description='The thought process of the agent in formulating the response.')
+ final_response: str = Field(..., description='The markdown-enabled response from the chat agent.')
+
+ class Config:
+ title = 'BasicChatAgentResponse'
+ description = 'Response from the basic chat agent. The response can be in markdown format.'
+ json_schema_extra = {
+ 'title': title,
+ 'description': description
+ }
+
+ system_prompt = SystemPrompt(
+ background=[
+ 'You are an expert in creating viral video ideas.',
+ 'Your goal is to generate engaging and creative video concepts that have the potential to go viral on social media platforms.'
+ ],
+ steps=[
+ 'Take the input given as the topic for the video ideas.',
+ 'Brainstorm and create a list of 10 unique and creative video ideas based on the given topic.',
+ 'Ensure that each idea is designed to capture attention and encourage sharing.'
+ ],
+ output_instructions=[
+ 'Ensure each idea is clearly numbered from 1 to 10.',
+ 'Output the list in human-readable Markdown format.',
+ 'Do not include any additional text or confirmation messages; only provide the list of 10 viral video ideas.'
+ ]
+ )
+
+
+ memory = ChatMemory()
+ initial_memory = [
+ {'role': 'assistant', 'content': 'Hello! I\'m an AI assistant that can help you generate viral video ideas. Please provide me with a topic to get started.'}
+ ]
+ memory.load_from_dict_list(initial_memory)
+
+ # Define dynamic info providers
+ def current_date_provider(format: str = '%Y-%m-%dT%H:%M:%S') -> str:
+ return f'The current date, in the format "{format}", is {datetime.now().strftime(format)}'
+
+ dynamic_info_providers = [DynamicInfoProvider(current_date_provider, 'Current date', format='%Y-%m-%d %H:%M:%S')]
+
+ agent = BasicChatAgent(
+ client=client,
+ model=model,
+ system_prompt=system_prompt,
+ memory=memory,
+ dynamic_info_providers=dynamic_info_providers,
+ output_schema=MyChatAgentResponse
+ )
+ console.print(f'Agent: {initial_memory[0]["content"]}')
+
+ while True:
+ user_input = input('You: ')
+ if user_input.lower() in ['exit', 'quit']:
+ print('Exiting chat...')
+ break
+
+ response = agent.run(user_input)
+ console.print(f'Agent: {response.final_response}')
\ No newline at end of file