noless 0.2.1a2

Multi-agent CLI for automatic dataset discovery and ML project generation

Noless CLI

NoLess: Multi-Agent AI Model Builder

                        ╔══════════════════════════════════════════════════════════════════════╗
                        ║                                                                      ║
                        ║  ███╗   ██╗ ██████╗     ██╗     ███████╗███████╗███████╗             ║
                        ║  ████╗  ██║██╔═══██╗    ██║     ██╔════╝██╔════╝██╔════╝             ║
                        ║  ██╔██╗ ██║██║   ██║    ██║     █████╗  ███████╗███████╗             ║
                        ║  ██║╚██╗██║██║   ██║    ██║     ██╔══╝  ╚════██║╚════██║             ║
                        ║  ██║ ╚████║╚██████╔╝    ███████╗███████╗███████║███████║             ║
                        ║  ╚═╝  ╚═══╝ ╚═════╝     ╚══════╝╚══════╝╚══════╝╚══════╝             ║
                        ║                                                                      ║
                        ║  Multi-Agent AI Model Builder | LLM-Powered Intelligence              ║
                        ║  Build AI Models Without Limits | Six Specialized Agents              ║
                        ║  Real-Time Code Generation | Intelligent Dataset Discovery             ║
                        ║                                                                      ║
                        ╚══════════════════════════════════════════════════════════════════════╝

NoLess is an advanced CLI-driven AI engineering system that uses a coordinated multi-agent architecture to automatically build machine learning projects from end to end. It searches datasets, designs architectures, generates production-ready code, manages training, and optimizes performance — all autonomously.

This approach eliminates boilerplate work and dramatically accelerates machine learning development.

---

Key Features

Multi-Agent Architecture

NoLess uses six specialized AI agents that collaborate to generate complete ML solutions:

Agent	Function
Orchestrator Agent	Controls workflow and execution
Dataset Agent	Searches OpenML, Hugging Face, Kaggle, UCI
Model Agent	Designs optimized architectures
Code Agent	Generates clean, production-ready code
Training Agent	Builds and manages the training pipeline
Optimization Agent	Performs hyperparameter tuning and improvements

Dataset Search

• OpenML (20,000+ datasets)
• HuggingFace Datasets Hub
• UCI Repository
• Kaggle Repository
• Task-aware filtering and dataset ranking

Real-Time Code Generation

Automatically generates:

• Model architectures
• Training scripts
• Preprocessing pipelines
• Evaluation metrics
• Configuration files
• Documentation

All files follow industry best practices and production-level standards.

Interactive CLI

• Step-by-step workflow
• Intelligent recommendations
• Rich outputs and enhanced usability
• ASCII banner and clean interface

Framework Support

• PyTorch
• TensorFlow / Keras
• scikit-learn

---

Installation

Option 1: Install from PyPI (Recommended)

pip install noless

Enable optional capabilities

The base install keeps dependencies minimal. Add feature sets on demand:

# Dataset downloader helpers (Hugging Face, Kaggle)
pip install "noless[data]"

# Training script generators / local experiments (PyTorch, TensorFlow, scikit-learn)
pip install "noless[ml]"

# LLM autopilot integrations (OpenAI, Anthropic)
pip install "noless[llm]"

# Everything
pip install "noless[all]"

Option 2: Install from Source

git clone https://github.com/DWE-CLOUD/NoLess.git
cd NoLess
pip install -r requirements.txt
pip install -e .

Verify Installation

noless --help

---

Quick Start

Interactive Mode

python -m noless.cli interactive

Direct Creation

python -m noless.cli create \
  --task image-classification \
  --framework pytorch \
  --agents

---

Usage Examples

Multi-Agent Project Creation

python -m noless.cli create \
  --task image-classification \
  --framework pytorch \
  --output ./my_classifier \
  --agents

Dataset Search

python -m noless.cli search \
  --query "diabetes classification" \
  --source openml \
  --limit 10

Dataset Download

python -m noless.cli download openml:37 --output ./data

---

Autopilot Mode (Ollama-LMM Powered)

NoLess can use local LLMs (via Ollama) to automatically plan, design, and build entire projects.

python -m noless.cli autopilot \
  --description "detect defects in solar panel images" \
  --output ./solar_inspector

Specify a model:

python -m noless.cli autopilot --llm-model deepseek-r1:7b

Autopilot performs requirement analysis, dataset extraction, dataset selection, downloading, multi-agent generation, and documentation creation.

---

Generated Project Structure

my_model/
├── train.py
├── model.py
├── config.yaml
├── utils.py
├── requirements.txt
└── README.md

All modules are cleanly structured, modular, and fully customizable.

---

Supported Tasks

Task	Description	Frameworks
Image Classification	Vision-based categorization	PyTorch, TensorFlow
Text Classification	NLP classification tasks	PyTorch, TensorFlow
Object Detection	Bounding box detection	PyTorch
Sentiment Analysis	Polarity scoring	PyTorch, TensorFlow
Regression	Numerical prediction	All
Clustering	Unsupervised grouping	scikit-learn
Time-Series Forecasting	Sequential prediction	PyTorch, TensorFlow
General NLP Tasks	Sequence and token tasks	PyTorch, TensorFlow

---

Multi-Agent Architecture

How It Works

1. The Orchestrator interprets the request
2. The Dataset Agent performs multi-source dataset search
3. The Model Agent creates an appropriate architecture
4. The Code Agent generates the necessary modules
5. The Training Agent constructs training workflows
6. The Optimization Agent tunes configurations and parameters

Communication

• Asynchronous message passing
• Shared context memory
• Priority scheduling
• Real-time updates

---

CLI Reference

noless search -q "query"
noless create -t TASK -f FRAMEWORK [--agents]
noless interactive
noless autopilot
noless download DATASET_ID
noless agents
noless templates

---

Configuration Example

task: image-classification
framework: pytorch

model:
  architecture: resnet50
  pretrained: true
  num_classes: 10

training:
  epochs: 50
  batch_size: 32
  learning_rate: 0.001

---

Roadmap

• Distributed training
• Automated model deployment
• Experiment tracking and model registry
• Additional dataset sources
• Web-based UI
• Custom agent plugins
• AutoML-style pipeline search

---

License

MIT License. Refer to the LICENSE file.

---

Acknowledgments

• OpenML
• Hugging Face
• PyTorch and TensorFlow teams
• Rich library
• Click CLI framework

---

Just tell me.