Scaling physical AI in days/hours.
Palette is the industry's first agentic software suite for building physical AI in days and hours, not months. Bring any model, compose the workflow, and move to SiMa.ai hardware with less friction.
SiMa.ai Palette Software Suite editor
EXPLORER
▾ Compile Models
├ model_compile.py
├ calibration
└ quantize
▾ Run Python App
└ app.py
MODEL COMPILE
Palette SDK compiles any ONNX model directly to SiMa.ai's MLSoC binary.
- Supports CNN, ViT, and GenAI model families.
- Mixed-precision quantization
PALETTE
Ask Palette anything:
- > Optimize this pipeline
- > Explain model latency
- > Deploy to target
PALETTE 3.0 NOW AVAILABLE
ANY Physical AI Model. REAL Machine Autonomy.
- Agentic ISP
- BEV kernels
- Camera–LiDAR
- 60 channels
3X greater perception, 15X faster camera tuning
Ultra-High-Density AI Video Processing: Process up to 60 independent 720p streams on a single Modalix MLSoC or 120 streams on two PCIe HHHL Cards
Agentic ISP tuning. Tune cameras without recompiling
Optimized BEV kernels: Resolved primary computational bottlenecks in 3D perception pipelines.
- Single chip
- Sensor fusion
- Neural depth
- Digital twin
More models, greater throughput
Optimized concurrent execution of Physical AI models on a single chip with interleaved execution and an optimized driver stack.
Docker Support on Modalix Run multiple dockers on Modalix to facilitate concurrent execution of AI models.
Unified sensor fusion, dense neural depth fields and digital twin simulation for drones and robotics
- EAGLE3
- DFlash
- Gemma 4 MTP
- MoE · GPT-OSS-20B
10X performance per watt within SWaP constraints
Speculative decoding with EAGLE3, DFlash and Gemma 4 multi-token prediction. White paper
Expanded Mixture-of-Experts support, including GPT-OSS-20B on 16GB Modalix. Explore these models at our Model Browser.
Smarter, more efficient AMRs with Integrated AI compute and software. White paper
Months of work.
Collapsed to days/hours.
Our software platform, Palette, is an agentic environment that collapses complex application timelines from months to days. The open source, purpose-built environment combines an execution library and agent workflow layer for fast development.
Seamless Migration
Seamless migration in a single prompt
Agentic Development
Build a conversational AI pipeline in minutes
Autonomous Deployment
Low latency, on device object detection built and deployed in minutes
How it all works.
From ONNX model to production binary on SiMa.ai's Modalix MLSoC — four steps, no surprises.
Coding Agent and Dev Tools
Developers (human + agentic) write, debug, and orchestrate projects.
Palette Agentic Environment
Containerized environment that builds, optimizes, and packages models for the Modalix MLSoC.
Palette Library
Runtime that orchestrates inference on both host PC and the MLSoC.
Modalix MLSoC
Performance
Browse, compare, and ship any model.
From classic CNNs to the latest GenAI models, SiMa.ai is built for what's next.
Model BrowserReady to build on SiMa.ai silicon?
Talk to our hardware team about your deployment requirements. Volume pricing, custom configurations, and evaluation kits available.
1 of 3
DLA-34
IMAGE CLASSIFICATION
Gemma 4 E2B
TEXT GENERATION · VISUAL QUESTION ANSWERING
Gemma 4 E2B Text
TEXT GENERATION