ABOUT SIMPLEX

AI models for robots
and real-time agents.

We build models, real-time agents and infrastructure for general-purpose robotic intelligence.

OUR DIRECTION

Connect understanding with action.

Our work brings interaction, navigation and manipulation into a unified streaming embodied architecture. We adapt multimodal foundation models for low-latency physical interaction, coordinating perception, speech and motion at different execution rates.

SA0 explores agents that talk, act and think concurrently. Social perception and multimodal memory provide context, while event handling and interruption recovery keep interaction moving. FastCD and VoiceLab make parts of this work available as developer and creator tools.

BUILT FOR THE PHYSICAL WORLD

One system. A connected world.

Simplex connects models with the infrastructure that makes them useful: real-time interaction, voice creation, structured decisions, perception and memory. We build across the model, runtime and product so these capabilities can work together.

Model researchRealtime infrastructureProducts that work together
01

Build for reality

Start with a real problem and follow the entire path to a useful result.

02

Make decisions legible

Expose structured outputs and candidate distributions, so builders can inspect what happened.

03

Learn in the open

Document research, engineering choices, and the limits of what has been verified.

Curious about the same questions?

Get in touch ↗