Description
SiMa.ai Palette is the software half of SiMa.ai's physical AI platform, the tooling used to develop and deploy applications on the company's Modalix MLSoC hardware. Where Modalix is the chip that runs models at the edge, Palette is the environment developers use to get a model onto that chip and turn it into a working application, in days rather than months, according to the company.
Read More
At the center of Palette is Palette Neat, an open-source agentic environment distributed as a container (ghcr.io/sima-neat/elxr) that combines a model compiler, an execution runtime and an agent workflow layer. Developers, human or AI, write and orchestrate projects with coding agents such as Codex or Claude inside VS Code or Palette's own Insight UI and command-line tool (sima-cli). Palette Neat then compiles and optimizes the model for the Modalix MLSoC through its model compiler and toolchain, and its runtime orchestrates inference across the host PC and the MLSoC device, with support for RTSP/NFS streaming and observability. SiMa.ai frames the workflow in four stages: developing with coding agents, compiling through Palette Neat, running the model through its runtime library, and finally executing on Modalix silicon.
SiMa.ai markets Palette for teams building physical AI applications: conversational AI running fully on-device, real-time object detection, and other perception and control workloads, across industries the company lists as drones, robotics, automotive, smart vision, industrial, government and healthcare. A Model Browser on SiMa.ai's site lets developers check model support before building, and the company offers a Modalix DevKit (50 TOPS, under 10W) for evaluating the hardware alongside Palette. The site directs teams that want volume pricing or custom configurations to contact its hardware sales team rather than listing self-serve prices; no public pricing for Palette itself appears on the site.
SiMa.ai was founded in 2018 and is headquartered in San Jose, California, with additional offices in Germany, India, Israel and Japan, on the premise that AI at the edge, rather than in the cloud, is where physical AI applications need to run.
- CategoryMLOps & Model Deployment
- Pricing ModelContact sales
- DeploymentCloud
- PlatformsWeb
SiMa.ai Palette's Core Features
Palette Neat agentic environment for compiling and packaging models onto the Modalix MLSoC
Built-in support for coding agents (Codex, Claude) inside VS Code or Palette's Insight UI
Command-line tool (sima-cli) for scripting builds and deployments
Model compiler and toolchain that optimizes ONNX models for the Modalix MLSoC target
Runtime APIs that orchestrate inference across the host PC and the on-device MLSoC
RTSP/NFS streaming support plus observability tooling for deployed models
Model Browser for checking model and workload support before development
Modalix DevKit hardware for evaluating the MLSoC before committing to production
How to use SiMa.ai Palette?
Explore models: use SiMa.ai's Model Browser to check whether your target model or workload is supported
Get the DevKit: order a Modalix DevKit to evaluate the MLSoC in real hardware
Develop: open a project in the Palette Neat container environment and work with a coding agent (or write code directly) in VS Code or the Insight UI
Compile: run the model through Palette Neat's model compiler and toolchain to target the Modalix MLSoC
Deploy and run: use the Palette Neat runtime to orchestrate inference across the host PC and the on-device MLSoC, then monitor it with the built-in observability tools
SiMa.ai Palette's Use Cases
- On-device conversational AI
- Real-time object detection
- Model evaluation before deployment
- Edge robotics and drone perception
