setpoints.ai
Observability for robots, world models, and physical AI agents.
Observability is now table stakes for software, cloud, and AI workloads1: servers, infrastructure, applications, and model systems already have mature platforms for logs, metrics, traces, and incidents.
The next frontier is physical. AI is moving from screens into the world: embodied agents, humanoid robots2, autonomous systems, learned world models3, and model-predictive control loops that plan over imagined futures before acting in reality.
setpoints.ai is building native observability for robotics and physical AI from simulation to deployment, starting with a trace data model designed from the ground up for embodied systems.
Join the setpoints.ai early access waitlist
No spam. Just early access when there is something real to show.
The platform
Debug the decision loop behind physical AI agents.
- SDKs - Python, C++, and Rust collectors for robot learning experiments, reinforcement learning runs, simulations, autonomy stacks, and deployed physical AI agents.
- Robotics Integrations - Native ingestion paths for MuJoCo, Isaac Lab, Gymnasium, Genesis, ROS 2, MCAP, Hugging Face, and LeRobot workflows.
- Physical AI Primitives - Typed trace events for observations, sensor frames, latent embeddings, predicted future states, action candidates, planned trajectories, executed commands, rewards, and real outcomes.
- World Model Debugging - Step through the decision loop, compare predicted latent rollouts against actual observations, and find exactly where a model of the world diverged from reality.
Stealth. Building in the open soon.