AISiMa.ai Modalix

Palette

Software

[SiMa.ai]

Palette is SiMa.ai's software suite for building and deploying Physical AI applications on the Modalix MLSoC: develop with Palette Neat, an agentic development environment; optimize models with the Model Compiler; choose them in the Model Browser; and deploy with the Device Manager.

System Overview
  • Type

    Software suite

  • Version

    Palette 3.0

  • Model Input

    ONNX

  • Target

    Modalix MLSoC

Product Information
  • Model

    Palette

  • Vendor

    SiMa.ai

  • Category

    On-Premise AI

  • Product Group

    SiMa.ai Modalix

  • Type

    Software

AceTec Support
  • Consulting

    Requirements analysis and custom system design

  • Implementation

    Integration guidance and lifecycle technical support

  • Verification

    Testing against specifications and industry standards

Features

  • Palette 3.0, announced by SiMa.ai on November 17, 2026
  • Develop: Palette Neat, an open-source agentic development environment
  • Optimize: Model Compiler for DNN and GenAI models, with mixed precision (INT8 and BF16)
  • Catalog: Model Browser with accuracy metrics, benchmarks and KPIs for each model
  • Execute: Device Manager for packaging, secure deployment and over-the-air updates
  • ONNX models compiled for the MLSoC; CNN, ViT and GenAI model families
  • Several models and Docker containers run at once on one Modalix MLSoC
  • LogCat, PerfCat and multi-core GDB debugging
  • Development on Linux, Windows and Mac

Overview

From model to application

Palette compiles an ONNX model for the Modalix MLSoC, with mixed-precision quantization, in a containerized environment that builds, optimizes and packages models; its runtime library orchestrates inference on the host PC and the MLSoC. Applications are written with SiMa.ai's Python APIs and frameworks such as OpenCV.

Palette Neat

In Palette Neat, developers and coding agents write, debug and orchestrate projects together, while it configures model requirements and orchestrates SiMa.ai's hardware in the background. Its repository, examples and reusable skills are open source.

Palette 3.0

Announced on November 17, 2026: concurrent execution of several models on one Modalix MLSoC, including separate Docker containers on the chip; speculative decoding with EAGLE3, DFlash and Gemma 4 MTP; and Mixture-of-Experts models such as GPT-OSS-20B on the 16 GB Modalix SoM.

Deployment

The Device Manager verifies, packages and securely deploys an application with its resources to Modalix boards in one workflow, and handles secure communication and over-the-air updates.

Specifications

Software

Item

Content

Suite

Palette 3.0, with Palette Neat

Flow

Develop (Palette Neat) · Optimize (Model Compiler) · Catalog (Model Browser) · Execute (Device Manager)

Model Input

ONNX

Model Families

CNN, ViT, LLM, LMM, GenAI

Precision

Mixed precision: INT8 and BF16

Applications

Python APIs, OpenCV

Debugging

LogCat, PerfCat, multi-core GDB, pipeline visualization (Insight)

Deployment

Device Manager: secure deployment, over-the-air updates

Development OS

Linux, Windows, Mac

Source

Open source on GitHub

Target

Modalix MLSoC: SoM, DevKit, PCIe card

Resources