Specialized Library For Non-LLM AI Agent Framework for embedded system

A Specialized Library, AbstractIntegratedModule for Non-LLM AI Agent Framework.

What is it?

AbstractIntegratedModule is an Open-sourced library specifically designed to provide complete NonLLM AI agent framework that lets you run multiple small AI agents that communicate peer-to-peer, share predictions, and learn collaboratively - all without a central server, a single Agent is a server provider and an active listener.

Why We built this

Most and many Non-LLM AI Agent systems are still centralized (LLM Wrapper based) - one model, one server, one point of failure. We wanted agents that could:

1. 🤝 Collaborate - Share predictions to peers when uncertain
2. 🔒 Work offline - No cloud dependency
3. 🧠 Learn continuously - From user feedback (Supervised and unsupervised learning)
4. 🌐 Scale naturally - Add more agents = more capacity for federated learning behavior to emerge gradually.

Key Features

Feature of What the library contains

Specialized MLP

A Regular MLP Implemented using Novel weight shaping based on data geometry (research-grade), proven effective on mid-hard synthetic samples, such as make_classification samples to real samples such as activity log

Transformer + MLP Ensemble

Combines both models for better accuracy, The Transformer is highly optimized with Alpha-based computation algorithm specifically for embedded systems.

Peer-to-Peer Inference

Agents consult each other when confidence is low, comparing their predictions with other peers.

Explainable AI

Shows WHY it made each prediction

Built-in Security

Authentication, rate limiting, trust levels, bootstrap authentication

Cross-session Memory

Remembers across restarts, the memory is saved to a single SQLite database, the memory can be exported or imported.

[=] A Snippet of our library usage, simple, efficient, with this setup, you could immediately have a predicted label without engineering MLP from scratch or via Pytorch.

```python

main_model = IntegratedPipeline(memory_name, use_async=True, ssl_cert_file=cert_file, ssl_key_file=key_file) # provide cert_file path or key_file path (optional)

main_prediction = PipelinePredictionManager(main_model, label_csv='example_manual_training.txt', target_title='window_title', label='label')

# example_manual_training is a .txt file that contain csv format like above example.

titles, y, label_map = main_prediction.load_labels_from_csv(<your_filename>, <target_title>, <target_label>)

# small training with simple titles

main_model.train(titles, y)

results, chosen_label, confidence = main_prediction.advanced_prediction_method(test_titles, label_map, example_rules,

show_proba=False, top_k=3,

use_transformer=True,

return_attention=False,

save_results=True)

# ... more features you can add

What makes it different

vs Traditional ML vs Centralized API vs Other P2P systems

Agents collaborate

No server costs, both agents is server and listener

Built-in security

Explains decisions

Works offline

Specialized ML for edge cases

Learns from feedback

Privacy-focused

Production-ready

Technical Highlights

1. 🐍 Pure Python - Works on any OS (Windows, macOS, Linux, ARM64)
2. 📦 Binaries provided - For efficient testing where pip is not supported.
3. 📦 One-line install - pip install AbstractIntegratedModule
4. 🔐 Enterprise security - API keys, rate limiting, trust levels
5. 🚀 Production ready - Officially proven works efficiently WITH Proofs in ARM64 Environment

Current Status

✅ Core P2P working

✅ Specialized MLP implemented

✅ Async message queue

✅ Security layer

✅ Cross-platform (x86, ARM64)

Roadmap

1. 🔜 Web UI dashboard
2. 🔜 More model architectures
3. 🔜 Robust and more Effective Federated learning support
4. 🔜 Wider adoption for Multi-agent research.

Who is this for?

1. Developers building a research to simple multi-agent systems
2. Researchers experimenting with collaborative Non-LLM AI for embedded systems.
3. Edge deployments where cloud isn't an option, and require tasks automation.
4. Privacy-focused applications keeping data local as a single SQLite database.
5. Collaborators: Who are interested and wants to improve the library Source code for Better deployment.

Links

🔗 GitHub: github.com/Micro-Novelty/IntegratedPipeline-Specialized-Non-LLM-AI-Agent-Framework

📦 PyPI: pip install AbstractIntegratedModule