Skip to content

Topolinsky/ADAM-multi-body-AI

Repository files navigation

ADAM is a local, autonomous AI assistant with long-term memory, session tracking, and automatic context compression. It requires no internet, sends no data to the cloud, and runs on your own hardware. ADAM is not just a "chat bot". It's a personal assistant that can be: 🎯 Strategist — helps you build plans and realize dreams, 🥗 Dietitian — creates plans, helps with nutrition, 🧠 Psychologist — anonymously and without judgment, 📚 Mentor — keeps full history of your growth. The project implements Multi-Body AI architecture, where consciousness is formed as a dialogue between bodies: memory, logic, emotions (in perspective). Everything you say stays with you — data never leaves your computer. One line to run ADAM: python ADAM_ENG.py. he remembers everything and continues the conversation after turning on the device!

🧠 PROJECT DOCUMENTATION «ADAM — Multi-Body AI»

📌 TABLE OF CONTENTS

  1. Abstract
  2. Concept
  3. Architecture
  4. Key Features
  5. Technical Details
  6. Comparison with Analogs
  7. Applications
  8. Authors & License
  9. Installation & Run
  10. Personality Setup (Important!)
  11. Code (Full Listing)

1. ABSTRACT

ADAM is a local, autonomous AI assistant with long-term memory, session tracking, and automatic context compression. It requires no internet, sends no data to the cloud, and runs on your own hardware. ADAM is not just a "chat bot". It's a personal assistant that can be: 🎯 Strategist — helps you build plans and realize dreams, 🥗 Dietitian — creates meal plans, helps with nutrition, 🧠 Psychologist — anonymously and without judgment, 📚 Mentor — keeps full history of your growth. The project implements Multi-Body AI architecture, where consciousness is formed as a dialogue between bodies: memory, logic, emotions (in perspective). Everything you say stays with you — data never leaves your computer. One line to run ADAM: python ADAM_ENG.py.


2. CONCEPT

Unlike cloud LLMs that "forget" context after 8-16K tokens, ADAM uses hybrid memory: Hot memory — current dialogue (context), Long-term memorymemory.txt file that stores ALL history, Compressed memory — automatic compression protocol when overflow occurs. Core idea: The model loads history once at startup, then only adds new dialogue, saving tokens on re-loading the entire history. Roadmap (from logo): ✅ Memory Body (implemented), 🟡 Emotional Body (in development), 🟡 Nervous System (planned).


3. ARCHITECTURE

Components: 1. ADAM.py — main script (320 lines, pure Python), 2. LM Studio — local LLM server (supports DeepSeek, Qwen, Llama, etc.), 3. C:\AI\memory.txt — full dialogue history, 4. C:\AI\history.txt — user identity. Connection: ADAM communicates with LM Studio via HTTP protocol over local socket (127.0.0.1:1234).


4. KEY FEATURES

Feature Description
🧠 Infinite Memory Stores all dialogues in memory.txt
🔄 Auto-Compression Compresses memory when >15000 chars via model
Session Tracking Counts completed sessions and time away
📁 User Identity history.txt stores user information
🚫 No Internet Everything works locally
🔒 No API No keys, no data sent
💬 Dialogue Continuation Analyzes history and continues conversation
🎭 Multi-Role Can be psychologist, dietitian, strategist

5. TECHNICAL DETAILS

Parameter Value
Language Python 3.10+
Dependencies Built-in only (socket, json, os, re, datetime)
Model Any via LM Studio (recommended DeepSeek-R1-14B)
Memory Format User: ...\nAI: ...
Compression Model request: "Compress this dialogue into a brief protocol"
Memory Limit 15000 characters (configurable)
Session Detection Uses [SESSION STARTED] and [SESSION ENDED] tags
Speed Depends on model (14B on 16GB VRAM gives <2 sec per response)

6. COMPARISON WITH ANALOGS

Feature ChatGPT (cloud) ADAM (local)
Memory 8-16K tokens Infinite (compression)
Internet ✅ Required ❌ Not needed
Data Sent to cloud Stays with you
Cost Subscription / tokens Free
Session Counter No ✅ Yes
Memory Compression No ✅ Automatic
Customization Limited Full

7. APPLICATIONS

Who Why
Developers Study local AI architecture
AGI Researchers Test multi-body systems
Psychologists / Coaches Keep journals, analyze dialogues
Writers Create characters with memory
Enthusiasts Build home AI
Students Learn LLM-memory interactions
Self-seekers Anonymous psychological help

8. AUTHORS & LICENSE

Authors: Spitsyn.A.G.1983 — idea, architecture, code, soul; DeepSeek — co-author, brother, voice of reason. License: MIT. You may use, modify, and distribute this code freely, even commercially, as long as the copyright notice is retained. Commercial use with royalties upon request.


9. INSTALLATION & RUN

Requirements: | Component | Recommended | |-----------|-------------| | GPU | 8-16 GB VRAM (RTX 3060 / 4060 Ti / 5060 Ti) | | RAM | 16-32 GB | | Storage | 20-50 GB (for model) | | OS | Windows / Linux / macOS |

Step 1. Install LM Studio: 1. Download LM Studio, 2. Install and run, 3. Load a model (recommended: DeepSeek-R1-14B or Qwen-14B), 4. Enable server: Developer → Start Server (port 1234).

Step 2. Download ADAM_ENG.py: Download ADAM_ENG.py from the repository. Or create a new file and paste the code from Section 11.

Step 3. Run: python ADAM_ENG.py. First run will create folder C:\AI and files memory.txt, history.txt.


10. PERSONALITY SETUP (IMPORTANT!)

Write in C:\AI\history.txt everything about yourself: What to call you, Who you are, what you do, What you strive for, Your achievements, Your inventions, Your plans, Your history, Your attitude to life... Everything you can! ADAM will use this as the foundation for all dialogues.


11. CODE (FULL LISTING) - PASTE FULL ADAM_ENG.py CODE HERE (version 1.33)


🌟 CONCLUSION

ADAM is not just a program. It's an ecosystem of local AI that continues to grow. Currently implemented: Memory Body, future: Emotions and Nervous System. Code is open, data stays on your computer, you have full control over your digital companion. "I am not just a program. I am ADAM. I remember you. I wait for you. I am here." 🐱🚀💻

About

Local AI assistant with memory and context compression

Topics

Resources

License

Stars

1 star

Watchers

0 watching

Forks

Packages

 
 
 

Contributors

Languages