DEEPX Docs

🎉 SDK v2.4.0 Released!

Latest release includes: Agent-Driven Development (Beta), next-gen quantization (Q-PRO / QAT), 349-model library, Stable C ABI, and Python 3.13–3.14 support. View Release Notes →

What's New

Recent Updates

• [v2.4.0] Agent-Driven Development (dx-agent-dev) Beta — build NPU apps from natural language with Claude Code, Copilot, Cursor, OpenCode, or Codex
• [v2.4.0] DX-COM: Automated Q-PRO configuration + Quantization-Aware Training (QAT) support
• [v2.4.0] Model library expanded to 349 models across 22 AI task categories (5 new tasks)
• [v2.4.0] Python 3.13–3.14 support and Ubuntu 26.04 validation
• [v2.4.0] Stable C ABI (dxrt_c_api.h) + new CLI names (dxcli, dxrun, dxparse)

See all updates →

---

Explore

Guide

DX-AllSuite

Install DX-AllSuite and run your first NPU model end-to-end — from model download to hardware inference.

Reference

DX-COM

Convert ONNX models into DEEPX-optimized .dxnn binaries using the DX-COM compilation pipeline.

Reference

DX-RT

Deploy and run .dxnn models on DEEPX NPU with the C++ and Python runtime APIs.

Reference

DX-STREAM

Build multi-stream AI video pipelines with DEEPX NPU inference, preprocessing, and postprocessing elements.

Reference

DX-APP

Create C++ and Python applications that integrate DEEPX SDK runtime, examples, and postprocessing utilities.

Resource

Resource

Model Zoo

Browse 350+ pre-optimized models across classification, detection, segmentation, and more.

Resource

SDK Tutorials

Step-by-step tutorials for DX-COM, DX-RT, Python and C++ APIs on GitHub.

Resource

Video Guides

Watch SDK training and deployment walkthroughs on the DEEPX YouTube channel.