Skip to content

ydsgangge-ux/AGI-DPA

BranchesTags

Repository files navigation

AGI Cognitive Assistant

A desktop AI assistant that simulates human cognitive architecture
layered memory · emotional weighting · associative retrieval · personality growth

Python PyQt6 License Platform

English | 简体中文

Features

  • A/B Dual Architecture — Layer A (consciousness) has personality, emotions, judgment; Layer B (executor) calls LLM + tools
  • Hierarchical Memory — SQLite + vector retrieval, three-tier storage (outline/detail/fragment) + associative network + two-phase retrieval
  • User Profile — Gradually accumulates personality traits, detects anomalous behavior, identity verification
  • 28 Built-in Tools — File operations, system control, web search, browser automation, OCR, coding agent, Office I/O, stock info, news, AI image generation
  • AI Image Generation — Auto-generates character selfies & scenery using pollinations.ai (free, no API key); proactive periodic generation (~3h); chat bubble image display; personality-bound avatar prompts with AI auto-generation
  • Growth Engine — Personality drift + active learning + experiential cognition with deduplication & activity decay — the AGI evolves through conversation
  • Mobile Web Client — Built-in web server (FastAPI), chat from any phone browser, shares the same agent instance and memory as desktop
  • Proactive Conversation — AGI initiates topics autonomously; user replies are stored as complete memory chains (system→user→AI)
  • 12 LLM Providers — DeepSeek / OpenAI / Claude / Gemini / Groq / Qwen / Zhipu / Doubao / Kimi / Baidu / SparkDesk / Ollama (100% local)
  • Multi-language — Chinese / English / 日本語 / 한국어 / Español / العربية
  • Voice Synthesis — Microsoft Edge TTS, multiple voices
  • Face Recognition — Multi-engine (InsightFace / face_recognition / OpenCV), multi-user identity
  • Desktop Integration — System tray, global hotkeys, floating window, screenshot OCR, auto-start

Quick Start

Windows (recommended)

  1. Install Python 3.10+ from https://www.python.org/downloads/
    • Must check Add Python to PATH
  2. Double-click install.bat — installs all dependencies
  3. Double-click launch.bat — starts the app

That's it! Two clicks after Python is installed.

Linux / macOS

# 1. Make sure Python 3.10+ and pip are installed
# Ubuntu: sudo apt install python3 python3-pip
# macOS:   brew install python3

# 2. Install dependencies
chmod +x install.sh launch.sh
./install.sh

# 3. Launch
./launch.sh

Screenshots

Main Chat Tool Panel Settings
Main Chat Tool Panel Settings

Project Structure

agi_app/
├── main.py                  # Entry point (PyQt6 desktop app)
├── server.py                # Mobile web server (FastAPI, shares agent instance)
├── install.bat / install.sh # One-click install scripts
├── launch.bat / launch.sh   # Launch scripts
├── build.py                 # PyInstaller packaging script
├── requirements.txt         # Python dependencies
│
├── engine/                  # AGI Core Engine
│   ├── models.py            # Data models (personality/memory/emotion/modality)
│   ├── memory.py            # SQLite vector memory store (CRUD + decay)
│   ├── memory_manager.py    # Hierarchical retrieval (two-phase retrieval)
│   ├── association.py       # Memory association network (directed weighted graph)
│   ├── agent.py             # Layer A consciousness agent (perceive→memory→reason→tool→generate)
│   ├── executor.py          # Layer B tool execution loop (ReAct, max 8 steps)
│   ├── tools.py             # 28 tool functions
│   ├── image_gen.py         # AI image generation (pollinations.ai, selfie & scenery)
│   ├── coder.py             # Autonomous coding agent (write→run→fix loop)
│   ├── office_tools.py      # Office file tools (docx/xlsx/pptx/pdf)
│   ├── user_profile.py      # User profile (trait accumulation + anomaly detection)
│   ├── learner.py           # Growth engine (personality drift + active learning + cognition dedup/decay)
│   ├── auth.py              # Multi-user identity verification
│   ├── face_recognition_engine.py  # Face recognition (three-engine lazy-load)
│   ├── simlife_client.py     # SimLife state reader (HTTP + file fallback)
│   ├── llm_client.py        # LLM client (DeepSeek/OpenAI/Groq/Claude/Gemini/Ollama)
│   ├── tts_engine.py        # Voice synthesis (Edge TTS / pyttsx3)
│   └── i18n.py              # Internationalization (6 languages)
│
├── desktop/                 # Desktop System Layer
│   ├── config.py            # Config management, paths, QSS dark theme
│   ├── system.py            # System tray, global hotkeys, auto-start
│   └── screenshot.py        # Screenshot selector + OCR background thread
│
├── simlife/                 # SimLife Virtual Life Simulation
│   ├── backend/             # FastAPI backend (auto-starts with main app)
│   │   ├── main.py          # Server entry + API routes (port 8769)
│   │   ├── world_engine.py  # Scene engine (schedule + weather + holidays)
│   │   ├── event_engine.py  # Daily/random/scheduled event system
│   │   ├── mood_engine.py   # Mood calculation (scene + events + weather)
│   │   ├── npc_engine.py    # NPC activation and interaction
│   │   ├── weather.py       # Open-Meteo weather (free, no API key)
│   │   ├── generator.py     # LLM character/NPC card generation
│   │   └── holiday_calendar.py  # Chinese holidays + festivals
│   ├── frontend/            # Setup web UI (initial character creation)
│   ├── data/                # Runtime data (character, world state, events)
│   └── setup.py             # Standalone setup launcher
│
└── ui/                      # UI Layer (PyQt6)
    ├── main_window.py       # Main window (7 functional tabs)
    └── float_window.py      # Floating window (topmost, draggable, animated, proactive replies)

First-time Configuration

After launching, go to the Settings tab to configure:

Setting Description
LLM Provider DeepSeek / OpenAI / Groq / Claude / Gemini / Ollama
API Key Get one from your provider's website (Ollama needs none)
Hotkeys Customize wake and screenshot hotkeys
Language Chinese / English / Japanese / Korean / Spanish / Arabic

Supported LLM Providers

Provider API Key URL Notes
DeepSeek https://platform.deepseek.com Recommended, affordable
OpenAI https://platform.openai.com GPT-4o-mini etc
Groq https://console.groq.com Free tier available, fast
Claude https://console.anthropic.com Anthropic
Gemini https://aistudio.google.com Google
Ollama https://ollama.ai 100% local, no key needed

Tool calling: DeepSeek / OpenAI / Groq / Qwen / Zhipu / Doubao / Kimi / Baidu / SparkDesk use native function calling. Claude / Gemini / Ollama use ReAct prompt parsing (tool descriptions embedded in prompt, JSON output). All providers support real tool execution.

Architecture Overview

User Input
    │
    ▼
① Perception (LLM) → Emotion / task type / topic tags
    │
    ▼
② Two-phase Memory Retrieval
   Phase 1: Vector search outlines + associative ripple spread
   Phase 2: Pull details by outline direction
   + User profile (always injected)
    │
    ▼
③ Reasoning (LLM) → Decide tool usage, storage strategy
    │
    ├── Needs tools ──→ ④ Layer B tool loop (ReAct, max 8 steps)
    │
    ▼
⑤ Generate Response (LLM) → Personality-driven output
    │
    ▼
⑥ Store → Layered memory by importance / emotion
    │
    ▼
⑦ Background → User profile / growth engine / experiential cognition

Tool List (28)

Category Tools
File System read_file · write_file · list_directory · search_files · delete_file
Execution run_command · run_python
Network web_search (DuckDuckGo + Bing) · fetch_url · read_article (newspaper3k)
System screenshot · mouse_click · keyboard_type · open_application · get_system_info · read_clipboard · write_clipboard
Browser browser_action (Playwright)
Office create_word · create_excel · create_pptx · create_pdf · read_office_file
Finance get_stock_info · search_stock
News get_news · get_news_sources
Image generate_image (pollinations.ai, free, no API key)

All high-risk tools (run_command, run_python) require explicit user confirmation before execution.

Keyboard Shortcuts

Shortcut Action
Ctrl+Shift+Space Show / hide floating window
Ctrl+Shift+S Select area screenshot + OCR

Both hotkeys are customizable in Settings.

Optional Enhancements

These are installed automatically by install.bat/install.sh if possible:

# Voice synthesis (Microsoft Edge TTS, free)
pip install edge-tts

# Mobile web server (chat from phone browser)
pip install fastapi uvicorn PyJWT

# Office file I/O (Word/Excel/PPT/PDF)
pip install python-docx openpyxl python-pptx reportlab pdfplumber

# Semantic vectors (improves memory retrieval quality, ~500MB)
pip install sentence-transformers

# Face recognition (InsightFace engine, recommended)
pip install insightface onnxruntime opencv-python

# Browser automation
pip install playwright && playwright install chromium

# Article extraction (intelligent news/article parser)
pip install newspaper3k

# Finance tools (stock info & search)
pip install yfinance

# News tools (requires API key from newsapi.org)
pip install newsapi-python

Missing optional dependencies gracefully degrade — core features still work.

Packaging as Standalone Executable

pip install pyinstaller
python build.py windows   # → dist/AGI-Desktop.exe
python build.py linux     # → dist/AGI-Desktop

Data Storage

All data is stored in the user directory — project folder stays clean:

Platform Data Directory
Windows %APPDATA%\AGI-Desktop\
Linux/macOS ~/.agi-desktop/

Core files:

  • config.json — User settings (API keys, hotkeys, etc.)
  • personality.json — Personality configuration
  • memory.db — SQLite database (memory/associations/user profile/face/growth)

Troubleshooting

"Python is not installed" or "'python' is not recognized"

  1. Download Python from https://www.python.org/downloads/
  2. During installation, check "Add Python to PATH" (critical!)
  3. Re-open Command Prompt and run install.bat again

"No module named PyQt6"

Run install.bat again, or manually: pip install -r requirements.txt

Garbled text in console

Right-click console title bar → Properties → Font → select a font that supports your language.

Memory retrieval quality is poor

Install semantic vectors: pip install sentence-transformers

Ollama tool calls don't work

Ollama doesn't natively support function calling. Use DeepSeek API for full tool support.

Tech Stack

  • UI: PyQt6 (dark theme)
  • LLM: DeepSeek / OpenAI / Groq / Claude / Gemini / Ollama
  • Memory: SQLite + sentence-transformers (optional)
  • Mobile: FastAPI + Uvicorn + PyJWT
  • Voice: Edge TTS / pyttsx3
  • Face: InsightFace / face_recognition / OpenCV
  • Office: python-docx / openpyxl / python-pptx / reportlab / pdfplumber
  • Browser: Playwright (optional)
  • Finance: yfinance
  • Article: newspaper3k
  • Image: pollinations.ai (free AI image generation)

License

MIT

About

电子女友/男友，具有人格、情感与分层记忆的桌面 AGI私人 助手 · 27 个内置工具 · 支持 6 大 LLM · 人格成长引擎；A desktop AGI assistant with personality, emotions & hierarchical memory · 27 built-in tools · 6 LLM providers · growth engine.

Topics

python pyqt5 local-llm desktop-ai ai-companion memory-system personality-engine vrm-avatar

Resources

Readme

License

MIT license
Activity

Stars

5 stars

Watchers

0 watching

Forks

1 fork
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages