Technology

Full-Stack AI, Proven in Mass-Produced Products

Every Codsaxi product runs on the AIYonder technology stack — a device-cloud collaborative architecture where the device handles real-time perception and privacy-sensitive tasks, while the cloud handles complex reasoning and continuous evolution. Every capability below is already shipped in production hardware.

🔌 Device Side: AI OS + On-Device Model Matrix

  • Offline voice wakeup: a hundred-KB-class model with ultra-low-power always-on listening.
  • On-device ASR / TTS: streaming speech recognition and emotional voice synthesis, fully local.
  • Vision & SLAM: face, gesture and pet recognition with autonomous navigation.
  • Local intent & motion control: commands parsed on-device, millisecond-level action response.
  • Works offline, data never leaves the device.

☁️ Cloud Side: AIoT Platform + Agent System

  • Multi-agent orchestration: task planning, tool calling and long-term memory.
  • Character engine: personality, emotion and growth systems, pluggable across products.
  • Device management: multi-protocol access, remote control and fault diagnosis.
  • Dual-mode OTA: fully automatic silent upgrades + user-controlled upgrades.

🧠 Multi-Model Collaboration Engine

Full-pipeline inference: ASR → NLU → retrieval → generation → TTS, with local and cloud paths working together. We integrate major foundation models (Azure speech, ChatGPT, and leading Chinese LLM providers) under a per-device licensing model, while our self-developed differentiating models build the AI's "soul": long-term memory, voiceprint recognition, one-brain-multiple-bodies, and virtual world building.

🔩 Six Chip Platforms, Full Price Coverage

Rich, shipping product examples across six silicon platforms — Rockchip RK/RV, ASR, ESP32, Jieli AC791, Bluetrum AB5800 and China Mobile 307 — giving partners flexible BOM options from entry-level to premium.

🧪 Model Engineering Pipeline

Training → compression → distillation → porting → on-device real-time inference: a complete engineering pipeline that puts large-model capabilities into MCU/SoC-class hardware.

Core Experience Technologies

  • Emotion computing engine: multi-dimensional emotion analysis from dialogue, tone and expression.
  • Long-term memory: vector-database memory architecture that remembers what users said years ago.
  • Virtual-real fusion: one unified interaction protocol keeps the AI consistent across app, toy and web.
  • Personalized growth: every AI companion evolves a unique personality with its owner.
  • Privacy computing: end-to-end encryption; wakeup and vision stay on-device.

Want a technical deep-dive or evaluation samples? Email bd@aiyonder.com.