NVIDIA's Jim Fan open-sources CaP-X agentic robotics framework
Jim Fan announced CaP-X, an open-source framework for agentic robots that auto-synthesizes skill libraries for arms and humanoids.
Jim Fan announced the open-source release of CaP-X, a framework for what he calls "vibe agents" that operate in the physical world as robot arms and humanoids. According to Fan's post on X, CaP-X ships with perception APIs, actuation APIs, and the ability to auto-synthesize skill libraries for robotic hardware.
The release lands agentic AI — until recently confined mostly to browser and coding tasks — squarely into robotics. By exposing perception and actuation as APIs and letting agents compose their own skills, CaP-X lowers the integration barrier between LLM-driven planners and physical manipulators. For builders, an open-source stack means the patterns that emerged in software agents (tool use, planning loops, skill reuse) can be prototyped on real hardware without a proprietary robotics platform. For buyers evaluating embodied AI, it signals that the tooling layer for agentic robotics is beginning to consolidate in public.