Description

The Module LLM Kit is a modular offline AI inference kit built around the Module LLM and Module13.2 LLM Mate. It brings large-language-model voice interaction, data communication, and host connectivity into a stackable M5Stack module design.

Based on the AX630C SoC, the Module LLM Kit provides 3.2 TOPS INT8 NPU performance, 4GB LPDDR4 memory, and 32GB eMMC storage. It can run built-in KWS, ASR, LLM, and TTS functions locally without relying on a cloud service, helping privacy-sensitive and network-limited projects keep voice and text processing on the device.

Notice: Before connecting the FPC cable, lift the connector latch first. After inserting the cable fully, press the latch down to secure it.

Before connecting the Module LLM FPC cable, remove the device frame as recommended. The FPC connector is located under the SPK module, so remove the SPK module before operation.

Model replacement requires AXERA-specific model processing. Existing general models from the market cannot be used directly.

Features

• Module LLM Kit with Module LLM and Module13.2 LLM Mate for offline AI inference.
• AX630C dual-core Cortex-A53 SoC with 3.2 TOPS INT8 NPU acceleration and native Transformer support.
• 4GB LPDDR4 memory, including 1GB for user applications and 3GB for hardware acceleration.
• 32GB eMMC5.1 storage for the Ubuntu-based system, models, and application data.
• Built-in KWS, ASR, LLM, and TTS functions for local voice interaction.
• Parallel multi-model processing, onboard microphone, speaker, RGB status LEDs, and ADB debugging.
• Module13.2 LLM Mate adds M5-Bus power stacking, CH340N USB conversion, Type-C log output, Ethernet, and expansion pads.
• Development support for Arduino, UiFlow2, StackFlow, apt-based model updates, and OpenAI API-compatible workflows.

Specifications

Applications

• Offline voice assistants that need local speech recognition, language understanding, and speech output.
• Smart home controllers that continue operating when cloud access is unavailable.
• Interactive robots and embedded terminals using the Module LLM Kit for local AI dialogue.
• Privacy-focused AI gateways, education projects, and prototype devices for edge inference.

Product List

• 1 x Module LLM
• 1 x Module LLM Mate
• 2 x FPC-8P Wire

Other Content

The Module LLM Kit communicates with host devices through serial interfaces and can expose logs, Ethernet, and expansion signals through the Mate board. Use the StackFlow resources and connection guide to match the wiring, FPC connection, model update method, and host workflow before deployment.

Resources

Module LLM Kit product document

Module LLM schematic PDF

Module13.2 LLM Mate schematic PDF

AX630C processor databrief PDF

Module LLM Arduino quick start guide

Module LLM UiFlow2 quick start guide

Module LLM ADB, UART, and SSH connection guide

Module LLM software update guide

Module LLM image update guide

Module LLM Arduino library repository

Module LLM Arduino API reference

Module LLM UiFlow2 API reference

FAQ

Can this kit run AI features without cloud access?

Yes. The Module LLM Kit is designed for local KWS, ASR, LLM, and TTS processing, so supported voice and language workflows can run on the device.

Which model is preinstalled?

The module is listed with Qwen2.5-0.5B preinstalled, and the apt repository provides additional supported model packages.

Can any downloaded large language model be used directly?

No. The supported models use an AXERA-specific format and must be processed before they can run correctly on the hardware.