Calabi builds open infrastructure for collaborative embodied AI. Our first release is RTSM (Real-Time Spatio-Semantic Memory)—an open-source spatial memory engine that turns sensor streams into a persistent, queryable world state.

SLAM answers “where am I?” and detection answers “what is that?” RTSM adds the missing piece: “where did I last see that thing, what state is it in, and can an agent reliably act on it?” We fuse metric geometry (SLAM backends like RTAB-Map / ORB-SLAM3) with open-vocabulary semantics (e.g., CLIP, segmentation) so robots and agents can query the world in human terms and receive grounded coordinates, history, and state transitions.

We’re research-first and building in public: publishing reproducible benchmarks, reference architectures, and developer tooling (visualizer, APIs). Near-term work includes YOLO-World integration and mobile ingest via ARKit odometry.

If you’re working on embodied agents, world models, robotics perception, or digital twins, we’d love collaborators and contributors.

Calabi Inc.

Org chart

Teams

Offices

Calabi Inc.

Org chart

Teams

Offices

Related companies