Exact Integer AI — Train, Inspect, Verify, and Deploy.
IGLnet is MLiglon’s browser-based demonstration laboratory for learning about, building, training, visualizing, analyzing, and exporting “fully-connected” Integer Gate Logic (IGL) neural networks. It uses deterministic exact-integer computation throughout, with flexible network topologies, multiple loss functions, configurable search strategies, Scale P values from 1 through 8192, detailed model inspection, experiment monitoring, checkpointing, benchmarking, and portable C export.
IGLnet includes robust CPU and WebGPU acceleration, including GPU-resident training, exact CPU/GPU verification, runtime shadow checking, resource diagnostics, and performance telemetry. Its continuously expanding Training Advisor analyzes dataset structure, model depth, search configuration, candidate quality, computational workload, GPU status, and training behavior to identify potential problems and recommend improvements. Together, these capabilities make IGLnet both an experimental research environment and a practical platform for exploring efficient, deterministic AI models for desktop, browser, Edge AI, and embedded deployment.