JOAT - Just One AI Tool

A multi-model conversation system that intelligently routes user queries to specialized AI models based on task type, maintaining conversation context for each model. Now with local Ollama models and a beautiful desktop GUI!

Features

Intelligent Task Classification: Automatically classifies user queries into different task types
Multi-Model Routing: Routes queries to the most appropriate AI model for each task
Conversation Context: Maintains separate conversation histories for each model
Local AI Models: Uses Ollama for cost-effective local AI processing
Beautiful Desktop GUI: Modern chat interface inspired by leading chat apps
Command Line Interface: Traditional CLI for power users
Easy Setup: Simple installation and configuration

🖥️ Desktop GUI vs Command Line

Desktop GUI (Recommended)

# On any OS (macOS, Windows, Linux)
python gui_app.py

# Or on macOS/Linux
./start_gui.sh

Features:

🎨 Beautiful modern interface
💬 Real-time chat experience
📊 Live model status display
🗑️ Easy conversation management
⌨️ Enter to send, Shift+Enter for new line

Command Line Interface

python main.py

Features:

⚡ Fast and lightweight
🔧 Full control and debugging
📝 Scriptable and automatable
🖥️ Works on any terminal

Supported Task Types and Models

JOAT uses a single comprehensive installation by default (regular profile). You can optionally switch to a small profile for lower resource usage. The mapping lives in models_mapping.json.

Regular profile models:

coding_generation: yi-coder:9b
text_generation: mistral:7b-instruct
mathematical_reasoning: qwen2-math:7b
commonsense_reasoning: mistral:7b-instruct
question_answering: llama3-chatqa:8b
dialogue_systems: qwen2.5:7b-instruct
summarization: llama3.2:3b-instruct
sentiment_analysis: llama3.1:8b-chat
visual_question_answering: llava:7b
video_question_answering: qwen2.5vl:7b

Small profile models (optional):

coding_generation: deepseek-coder:1.3b
text_generation: llama3.2:1b
mathematical_reasoning: deepscaler:1.5b
commonsense_reasoning: tinyllama:1.1b
question_answering: phi3.5:3.8b
dialogue_systems: qwen2.5:3b-instruct
summarization: phi3.5:3.8b
sentiment_analysis: llama3.2:1b
visual_question_answering: moondream:1.8b
video_question_answering: qwen2.5vl:3b

Each model in the profiles is selected to specialize per task (coding, math, QA, etc.), balancing quality and performance. You can switch profiles at runtime if needed.

🚀 Quick Start

1. Install Ollama

# macOS
brew install ollama

# Linux
curl -fsSL https://ollama.ai/install.sh | sh

# Windows
# Download from https://ollama.ai/

2. Start Ollama

brew services start ollama  # macOS
# or
ollama serve               # Any platform

3. Install Python Dependencies

pip install -r requirements.txt

4. Install AI Models (Default: Comprehensive)

python setup_comprehensive_models.py

5. Launch the App

Desktop GUI (Recommended):

python gui_app.py
# or on macOS/Linux
./start_gui.sh

Command Line:

python main.py

📦 Model Installation

The app will automatically download models as needed, but you can pre-install them:

Install Models

# Default comprehensive (regular profile) interactive installer
python setup_comprehensive_models.py

# Optional minimal one-shot installer (installs regular profile without prompting)
python setup_ollama.py

Manual Installation

# Pull any model from the mapping manually
ollama pull yi-coder:9b
ollama pull qwen2.5:7b-instruct
... # etc.

Profiles

small_sized_models: optimized for speed and low memory.
regular_sized_models: balanced quality and performance (default install).

🎯 Usage Examples

Desktop GUI

Launch: python gui_app.py (or ./start_gui.sh on macOS/Linux)
Type your query in the input box
Press Enter or click Send
Watch the AI route your query to the best model
See which model was used in the response

Command Line

python main.py

You: Write a Python function to calculate fibonacci numbers
🤖 [Response from codellama]

You: Solve the equation 2x + 5 = 13
🤖 [Response from wizard-math]

You: What is the capital of France?
🤖 [Response from llama3]

🔧 Configuration

Models Mapping Format

Edit models_mapping.json to customize which models handle which tasks per profile:

{
  "regular_sized_models": {
    "coding_generation": "yi-coder:9b",
    "text_generation": "mistral:7b-instruct",
    "mathematical_reasoning": "qwen2-math:7b",
    "commonsense_reasoning": "mistral:7b-instruct",
    "question_answering": "llama3-chatqa:8b",
    "dialogue_systems": "qwen2.5:7b-instruct",
    "summarization": "llama3.2:8b-instruct",
    "sentiment_analysis": "llama3.1:8b-chat",
    "visual_question_answering": "llava:7b",
    "video_question_answering": "qwen2-vl:7b-instruct"
  },
  "small_sized_models": {
    "coding_generation": "stablecode:3b",
    "text_generation": "llama3.2:1b",
    "mathematical_reasoning": "deepscaler:1.5b",
    "commonsense_reasoning": "tinyllama:1.1b",
    "question_answering": "phi3.5:3.8b",
    "dialogue_systems": "qwen2.5:3b-instruct",
    "summarization": "phi3.5:3.8b",
    "sentiment_analysis": "llama3.2:1b",
    "visual_question_answering": "moondream:2",
    "video_question_answering": "qwen2-vl:2b-instruct"
  }
}

Selecting a Profile at Runtime

Default: auto-detected. The app uses regular_sized_models only if all regular-profile models in models_mapping.json are installed. Otherwise, it uses small_sized_models.
Force via env var: export JOAT_PROFILE=small_sized_models or regular_sized_models
Programmatic override: JOATSystem(models_mapping_file="models_mapping.json", profile_key="small_sized_models")

🛠️ Troubleshooting

See the Advanced Guide for troubleshooting and performance tips.

📜 License

The JOAT source code is licensed under the Apache License 2.0. You can find the full license text in the LICENSE file.

AI Model Licenses

The AI models used by this project have their own licenses. When you use JOAT, you are also subject to the license terms of the underlying models, which include:

Llama 3 Community License: Applies to models like llama3 and codellama.
Apache 2.0: Applies to models like mistral.

It is your responsibility to review and comply with the terms of each model's license and its Acceptable Use Policy.

📁 File Structure

joat/
├── main.py              # CLI version
├── app.py               # Core application logic
├── gui_app.py           # The GUI application window
├── start_gui.sh         # GUI launch script for macOS/Linux
├── ollama_client.py     # Ollama API client
├── setup_ollama.py      # Model setup script (installs regular by default)
├── models_mapping.json  # Task-to-model mapping (profiles)
├── requirements.txt     # Python dependencies
├── docs/                # Documentation files
└── README.md            # This file

🎨 GUI Features

Modern Design: Clean, modern chat-like interface
Real-time Status: Live Ollama and model status
Model Information: Shows which models are available
Conversation History: Maintains chat history
Keyboard Shortcuts: Enter to send, Shift+Enter for new line
Responsive Layout: Adapts to window size
Error Handling: Graceful error messages

🔄 Conversation Management

In GUI:

Click "🗑️ Clear History" to clear all conversations
Each model maintains its own conversation context
Automatic model switching based on task type

In CLI:

Type clear to clear history
Type history to see conversation history
Type status to check system status
Type quit to exit

📈 Performance Tips

SSD Storage: Models load faster from SSD
RAM: 16GB+ recommended for smooth operation
GPU: Optional but speeds up inference (if supported by your hardware and Ollama)
Model Selection: Use smaller models for faster responses

🤝 Contributing

Fork the repository
Create a feature branch
Make your changes
Test thoroughly
Submit a pull request

Enjoy your local AI assistant! 🤖✨

Name		Name	Last commit message	Last commit date
Latest commit History 17 Commits
.idea		.idea
__pycache__		__pycache__
docs		docs
.DS_Store		.DS_Store
.gitattributes		.gitattributes
LICENSE		LICENSE
README.md		README.md
app.py		app.py
example.py		example.py
gui_app.py		gui_app.py
main.py		main.py
models_mapping.json		models_mapping.json
notes.txt		notes.txt
ollama_client.py		ollama_client.py
requirements.txt		requirements.txt
setup_comprehensive_models.py		setup_comprehensive_models.py
setup_ollama.py		setup_ollama.py
start_gui.sh		start_gui.sh
test_models.py		test_models.py

License

YoniBL/joat

Folders and files

Latest commit

History

Repository files navigation