KalEdge is a platform designed to bridge the gap between high-level Deep Learning model design and efficient hardware (FPGA) implementation. Powered by hls4ml, KalEdge adds layers of automation, AI-assisted optimization, and remote hardware synthesis workflows to make Edge AI accessible to software engineers and hardware designers alike.
Official Cloud Platform: kaledge.kaleidoforge.com
Documentation Portal: GitHub Repository
- Code-Free HLS Synthesis: Go from a Keras/QKeras model to a fully synthesizable C++ Vivado HLS project with optimized reuse factors and precision in just a few clicks.
- AI Architect: Leverage a session-local LLM agent (powered by Anthropic Claude under a Bring Your Own Key model) to design optimized hardware-ready architectures.
- Comprehensive Compression Suite:
- Pruning & Quantization Aware Training (QAT): Minimize weight density and bitwidths using TensorFlow Model Optimization Toolkit (TF-MOT) and QKeras.
- Quantization-Aware Pruning (QAP): Jointly prune and quantize weights simultaneously to maximize silicon efficiency without accuracy degradation.
- Knowledge Distillation (KD): Compress complex neural networks into ultra-lightweight "Student" models optimized for silicon.
- Low-Rank Factorization (SVD): Decompose dense layers to save resources.
- Preset Optimization Pipelines: Execute pre-established compression combinations with a single click.
- hls4ml Integration: Convert and export synthesizable HLS projects with optional AXI4-Stream / DMA IP wrapper support.
KalEdge follows an Open-Core distribution strategy:
| Tier | Distribution | Auth | Features |
|---|---|---|---|
| Community | Docker image (this repo) | Supabase account (free) | Full compression suite, AI Architect (BYOK), hls4ml export |
| Pro / Corp | Institutional delivery | Hardware-bound license | All Community features + FPGA estimators, Build Agent, AI Advisor |
The Community edition is distributed as a Docker image. It requires a free account at kaledge.kaleidoforge.com to log in.
docker pull ghcr.io/kaleidoforge/kaledge-community:latest
docker run -p 8080:8080 ghcr.io/kaleidoforge/kaledge-community:latestThen open your browser at http://localhost:8080 and log in with your KalEdge account.
- Register: Create your free account at kaledge.kaleidoforge.com.
- Run the image: Pull and run the Docker image as shown above.
- Upload Datasets: Upload your data in a supported
.csvformat (see Dataset Format Guide) or load native datasets (MNIST, CIFAR-10). - Configure AI Architect: Enter your Anthropic API Key in the sidebar to unlock conversational network design.
- Train & Compress: Establish your baseline accuracy, run knowledge distillation, prune redundant weights, and quantize the model down to custom bit precision.
- Export: Convert the model using the hls4ml tab and download your synthesizable HLS project.
kaledge/
├── Dockerfile # Community image build definition
├── LICENSE # AGPL-3.0 open-source license
├── README.md # Main entry point and platform overview
│
├── .github/workflows/ # CI/CD for automated Docker image publishing
├── docs/ # Technical manuals and guides
├── examples/ # Example datasets and notebooks
└── tools/ # Utility scripts
This repository is open-sourced under the GNU Affero General Public License v3.0 (AGPL-3.0).
This means you are free to use, modify, and distribute this software, but any modified version used to provide a network service must also be released under the same license.
Please note that this license does not apply to the KalEdge Pro/Corp platform, its proprietary estimators, build servers, or enterprise delivery assets, which remain closed-source commercial products of KaleidoForge. See the LICENSE file for complete scope details.
© 2025–2026 KaleidoForge | Lightning Bridges for Fast AI