MinimalLLMAgent 0.1.0

A minimal LLM agent with memory management.

Overview

Features:

• simple and unified
• memory management
• a terminal simulation that allows for web-style interaction

Supported platforms/LLMs:

• OpenAI
• Grok (xAI)
• DeepSeek
• Gemini (Google)
• Qwen (Alibaba)

Examples

See the demo folder.

1: String Input

from src.min_llm_agent import min_llm_agent_class

if __name__ == "__main__":

    llm_agent = min_llm_agent_class(platform_name="OpenAI", model_name="gpt-4o-mini")

    question = "What is the capital of France?"
    response = llm_agent(question)

    print(f"Question: {question}")
    print(f"Answer by model {llm_agent.model_name}: {response}")
    
    # llm_agent.print_memory(memory_item_separator="/")
    llm_agent.print_memory()

2: Dict Input

from src.min_llm_agent import min_llm_agent_class

if __name__ == "__main__":

    llm_agent = min_llm_agent_class(platform_name="OpenAI", model_name="gpt-4o-mini")

    messages = [
        {
            "role": "system",
            "content": "You are a helpful assistant.",
        },
        {
            "role": "user",
            "content": "What is the capital of France?",
        }
    ]
    response = llm_agent(messages)

    print(f"Messages: {messages}")
    print(f"Answer by model {llm_agent.model_name}: {response}")
    
    # llm_agent.print_memory(memory_item_separator="/")
    llm_agent.print_memory()

3: Interact

from src.min_llm_agent import min_llm_agent_class

if __name__ == "__main__":

    llm_agent = min_llm_agent_class(platform_name="OpenAI", model_name="gpt-4o-mini")
    llm_agent.interact()

==========================

This is Yue's minimal LLM agent, powered by the model "gpt-4o-mini".

- To submit a query: start a new line, type '/', and press Enter.
    - Line breaks are allowed and recognized as a part of the query.
- Query 'q' or 'Q' to exit.
- Query 'm' or 'M' to print the memory.

See more details on: https://github.com/YueLin301/min_llm_agent

>>>>>>>>>>>>>>>>>>>>>>>>>>
[0] Question:
> 1+1=
/

<<<<<<<<<<<<<<<<<<<<<<<<<<
[0] Answer by the model gpt-4o-mini:
1 + 1 = 2.


>>>>>>>>>>>>>>>>>>>>>>>>>>
[1] Question:
> how are you
/

<<<<<<<<<<<<<<<<<<<<<<<<<<
[1] Answer by the model gpt-4o-mini:
I'm just a computer program, so I don't have feelings, but I'm here and ready to help you! How can I assist you today?


>>>>>>>>>>>>>>>>>>>>>>>>>>
[2] Question:
> m
/
==========================
Memory:
--------------------------
[0] (user): 1+1=
[1] (assistant): 1 + 1 = 2.
[2] (user): how are you
[3] (assistant): I'm just a computer program, so I don't have feelings, but I'm here and ready to help you! How can I assist you today?
==========================
>>>>>>>>>>>>>>>>>>>>>>>>>>
[2] Question:
> q
/

How to Use

Installation

pip install 

API Key

For security reasons, this project does not maintain any API key files. You need to configure the API key yourself in the environment variables. Check the following guidelines to see how it is done:

• OpenAI guideline
• Alibaba guideline

Resources:

• LLM Comparisons
• OpenAI
  • Models and pricing
  • How to create API keys
  • Billing
• Grok
  • Models and pricing
  • API Guides
  • A query example
  • A query example - OpenAI lib
• DeepSeek
  • Models and pricing
  • API
  • Billing
• Gemini
  • Models and pricing
  • API
  • A query example
  • A query example - OpenAI lib
  • Billing
• Alibaba
  • Models and pricing
  • API Guides
  • A query example
  • A query example
  • Dashboard

An Example Set Sp for MacOS Users:

1. Append the following API configurations to the end of the ~/.zshrc file.

export OPENAI_API_KEY="sk-xxx"
export OPENAI_BASE_URL="https://api.openai.com/v1"

export XAI_API_KEY="xai-xxx"
export XAI_BASE_URL="https://api.x.ai/v1"

export DEEPSEEK_API_KEY="sk-xxx"
export DEEPSEEK_BASE_URL="https://api.deepseek.com"

export GEMINI_API_KEY=""
export GEMINI_BASE_URL="https://generativelanguage.googleapis.com/v1beta/openai/"

export DASHSCOPE_API_KEY="sk-xxx"
export DASHSCOPE_BASE_URL="https://dashscope.aliyuncs.com/compatible-mode/v1"

2. Run source ~/.zshrc to update.