SenseNova‑MARS is the first open‑source agentic vision‑language model that can see, reason, and act. Built on a hybrid transformer‑Mamba architecture, it offers low memory usage, fast inference, and built‑in policy controls for secure deployment.