Unlike conventional ChatGPT-style services or large language models that require massive power consumption, LLiMa enables heavy AI models to run smoothly on edge devices with extremely low power consumption of under 10 watts.

LLiMa supports immediate deployment of widely used open-source models, such as Meta’s Llama, as well as customer-customized models.

LLiMa supports not only text-based language models, but also camera-based vision models and multimodal models that combine both language and vision. All models can be run and managed within a single LLiMa environment, without the need to switch between multiple tools.

Rather than simply placing an AI model onto a device, Palette SDK allows developers to precisely design the entire workflow pipeline, from processing camera input to controlling devices, using C++ or Python.

The dedicated compiler automatically optimizes the developer’s code for SiMa.ai’s MLSoC hardware architecture, enabling maximum speed and performance while consuming minimal power.

Completed software can be reliably deployed not only on lab PCs, but also directly onto real-world edge devices such as robots, drones, and smart factory equipment. Palette SDK also includes Device Manager, a management tool that allows multiple devices to be controlled at once.