DXNN® — the DEEPX NPU SDK for compiling, optimizing, and deploying AI inference at the edge.
DEEPX builds high-efficiency Neural Processing Units (NPUs) and the full software toolchain around them. Our open-source stack, DXNN® (DX-AllSuite), streamlines the entire journey — from an ONNX or PyTorch model to a real-world "Physical AI" application running on a DEEPX NPU — through a single, version-aligned toolchain covering compilation, optimization, and deployment.
- High efficiency — the proprietary DX-COM compiler extracts full NPU performance with Intelligent Quantization (INT8) that minimizes accuracy loss.
- Seamless integration — build vision analytics pipelines with DX-Stream (GStreamer-based plugins) without heavy code changes.
- Flexible ecosystem — Python and C++ APIs, a ModelZoo of 497 optimized models, and workflows for popular frameworks (Ultralytics, PaddlePaddle, RapidAI).
Try it now
- Want to build an app in natural language? Check out dx-agent-dev (showcase) inside dx-all-suite — DEEPX Agent-Driven Development: describe an app in plain language and an AI coding agent builds it end to end for the DEEPX NPU.
- Want to experience the entire DXNN SDK in one browser? Open dx-ai-studio — a desktop web workspace with model catalog, compiler, inference, streaming, benchmarking, hardware monitor, and an agent-driven builder in a single screen.
🔗 Company homepage → deepx.ai
Everything you need to start building on the DEEPX NPU lives on the developer portal.
- developer.deepx.ai — developer portal home
- ModelZoo — 497 optimized, ready-to-run models
- DXNN Introduction (lecture & video) — SDK overview and getting-started videos
- DEEPX × Rapidflare AI Assistant — an AI-powered platform that makes DEEPX's technical knowledge (GitHub docs, DXNN SDK, Developer Portal, YouTube tutorials) queryable through one interface, returning instant answers cited with deep links
Right now, open builders are shipping edge-AI applications on DEEPX NPUs together — and these are the best places to start.
| Project | What it is |
|---|---|
| dx-all-suite | The all-in-one SDK integration & release environment that ties every DXNN component together with aligned versions. Start here. |
| dx-tutorials | Install guides for the SDK and key demos — the fastest way to get hands-on. |
| dx-runtime | Deployment platform for the target system — DX-RT, NPU driver/firmware, DX-APP (C++/Python examples), and DX-Stream. |
| dx-compiler | Host-side compiler platform — DX-COM converts ONNX models into NPU-ready .dxnn binaries. |
| dx-modelzoo | 497 optimized, ready-to-run models with evaluation via ONNX Runtime or DX-Runtime, plus retraining recipes. |
Browse all repositories → github.com/orgs/DEEPX-AI/repositories
- New to DEEPX? Follow the Get Started guide, or install the SDK and demos via dx-tutorials.
- Building an app? Explore the C++/Python examples in dx-app and the pipelines in dx-stream.
- Bringing your own model? Compile it with dx-compiler, or pick a starting point from the ModelZoo.
- Want to try natural-language app building? Check out dx-agent-dev inside dx-all-suite.
Visit developer.deepx.ai to learn more.