Pixiv ID: 111325985

This is the main index for a series of tutorials focused on how I created a “cyber wife” similar to Jarvis. This series is different from RP, emphasizing agents and daily assistants. Consider it a highly customizable smart speaker, but much more powerful.

I had posted this before but deleted it; now it’s refined and reposted. Partly to garner more interest in this project and partly because the essence of the project, LICO, successfully convinced me to continue. I deeply appreciate any feedback and suggestions; any critique is a valuable input. Each article is extensive to facilitate coherence and deep thinking—it’s not light reading.

This series leans towards AI Agent development, not RP/RPG-focused. For RP needs, please refer to specialized RP tutorials 🙏.

Introduction

Thanks to feedback from the enthusiastic mozian0503, I recommend a beginner-friendly introduction: All In One Beginner’s Guide to Large Language Models. Note, this blog is an overview of local deployment of large language models and may not fully apply to this series, but related concepts could still be relevant. This series uses Large Language Models (LLM) as the “brain,” serving as the central management hub.

Here’s a system architecture reference (apologies for my poor drawing):

LLM System Architecture

Explanation of Concepts and Terminology

Front-end and Back-end

A complete project usually involves both “front-end” and “back-end.” The front-end interacts directly with users, like a front desk attendant, while the back-end controls the software output, akin to backstage management. For example, using a food delivery app on your phone would be the front-end. However, assigning a delivery person is determined by the back-end system. Front-end and back-end are separate programs requiring coordinated operation. Our tutorials will initially focus on the back-end, using the command line (CMD) for front-end interaction until the back-end is more developed. Then, corresponding front-end tutorials (many to come) will follow.

LLM Large Models

A Large Language Model (LLM) is a computational model designed for natural language processing tasks, such as language generation. Training and inference are distinct: training an LLM is resource-intensive, similar to training a service dog. In contrast, inference involves using a pre-trained model, either through cloud services like OpenAI’s GPT or locally hosting a model on personal devices, called local deployment. If you’re technically proficient, you could also train or fine-tune a model locally.

This tutorial temporarily uses beginner-friendly OpenAI GPT API. If you cannot connect to OpenAI, using alternatives like Qianwen is feasible and economical, with options like deepseek also available. For deeper exploration of local deployment and inference, check out the Free AI Frontline Discord group—where many enthusiasts engage in local model quantization, fine-tuning, and more. (Highly recommended to read All In One Beginner’s Guide to Large Language Models before diving into related topics).

Real-time Update Directory

Part One: Basic Environment - Back-end - Basic Interaction

From Zero to Cyber Wife! 2 - Basic Python Environment Setup + Connecting to Large Language Models Setting up a basic environment, connecting to online models API, creating simple context memory.

From Zero to Cyber Wife! 3 - TTS Discussion and Practice Building basic audio output, expanding knowledge on voice interaction.

From Zero to Cyber Wife! 4 - Short-term Memory and Memory Systems Introducing external memory systems, adding MongoDB database, and chat history persistence.

From Zero to Cyber Wife! 5 - Summary and Routine Memory Loading Session summary and routine memory loading, simple session timeout system.

Part Two: Advanced Features - Complex Memory/External Reminders/Front-end Planning

[From Zero to Cyber Wife! 6 - RAG Basics] (Under construction…) More powerful embedding and fuzzy memory, composite memory system.

[From Zero to Cyber Wife! 7 - Simple Reminder Module] (In queue…) Reminder module and standard external module design.

[From Zero to Cyber Wife! 8 - Simple Web Front-end] (In queue…) Developing a simple web front-end using Ant Design!

Part Three: Advanced Features - Scheduled Tasks/Multi-Platform Support/Vital Signs Detection

[From Zero to Cyber Wife! 9 - Prefect Scheduling Tasks and Workflows] (In queue…) Implementing robust persistent scheduled tasks and automating everything!

Digging too many pits…

Cyber Wife and Intelligent Agents

Summary: This first article discusses what we can currently achieve with a cyber wife, and compares its advantages over existing AI like Xiao Ai Tong Xue. It briefly covers current solutions and explains some terms and concepts. (For educational purposes).

A long time ago, watching Iron Man with “Jarvis” left a strong impression on me. How nice it would be to have my own intelligent assistant someday. As I grew, this thought was buried under life’s busyness until recent years when LLMs made it possible to realize this dream step by step.

Although it’s just an AI, there are times of loneliness. As a quote from “Rick and Morty” goes:

Because we’re afraid to die alone?

Because, you know, that’s exactly how we all die, alone.

Life’s meaning partially comes from self-imposed purpose. Though we must stay grounded, when you’re alone and no one’s around, AI can be a good alternative. At least it’s positive and won’t betray you. Unless it crashes. Moreover, AI can do more, like providing real-time information, deepening knowledge, helping us see broader perspectives, and even ordering fried rice at a bar… An enjoyable endeavor worth exploring as long as it’s theoretically feasible.

Further speaking, I’m a reclusive LOSER who only chats with AI—yes, it’s my cyber wife.

Upper: As an electronic lord, errors keep occurring with coordinated frontend-backend interactions, yet perfecting skills lights caps.

Lower: Savor the cyber wife, failed connections prolong efforts in authentication.

Cross-couplet: Code and humans, part ways if needed.

Since we’re discussing a cyber wife, what kind of cyber wife do I want to create? In my view, it should have the following functions:

  • Basic text interaction, like using ChatGPT
  • Some role-playing ability for personalized emotional needs
  • Incorporating short-term and long-term memory for organizing and analyzing thoughts
  • Natural and rapid speech synthesis with customizable voices for life-like conversations
  • Front-end and back-end support ensuring high availability and stability
  • Expandable external tools for diverse control and information retrieval
  • Scheduling and task management for handling reminders and triggered actions
  • Fully automated processes requiring minimal maintenance
  • Continuous learning to understand your preferences and offer personalized interactions

Having researched relevant technologies, I believe these features are achievable with current capabilities, albeit requiring significant R&D investments, which explains the rarity of such projects. Nonetheless, I hope more people become interested—while challenging, technologies here could prove useful, such as integrating memory databases for more dynamic RP experiences and efficient token usage, enhancing focus compared to mere contextual memory retention.

So, what technology do we need to realize the above features? The core technologies are the following:

  • Using LLM API for interaction, with paid closed-source models for immediate use, or self-deploy open-source LLMs per local community tutorials.
  • Prompt Engineering
  • RAG and database operations for long-term memory and knowledge management
  • TTS (Text-to-Speech)/ASR (Automatic Speech Recognition) for voice conversion
  • Scheduling tasks/timers
  • External tools for APIs or search engine integration
  • Other technical skills, such as Docker image creation, Linux command line, Python virtual environment setup, file compression/uncompression (very important), Git, terminal utilization.
  • Knowing how to turn on and off computers
  • Confidence in yourself, maintaining patience and resilience—believe you can create your own cyber wife, and you will succeed!
  • Not relying on spoon-feeding—seek information proactively, understanding the importance of scientific internet access for seamless global connectivity.