We developed CRAYON, a tokenizer processing up to 20 million tokens per second, and Swades, an open-source, terminal-native autonomous coding agent.
Active Projects
A hardware-accelerated vocabulary swapping tokenizer written in C++17. It supports CUDA and HIP kernels, loads pre-built vocabulary profiles using zero-copy memory mapping (`mmap`), and achieves tokenization throughput of up to 20 million tokens per second.
A terminal-native autonomous coding agent built on the Reasoning + Acting (ReAct) loop pattern. It runs automated sub-tasks in parallel Git worktrees to isolate workspace edits and features native GUI automation support for Wayland/GNOME via Mutter D-Bus remote desktop portals.
Bottom Line Up Front: Connect with XERV for expert inquiries regarding ethical AI research, dataset licensing, or technical partnerships to advance high-performance Human-AI synergy. Our team provides direct support for integrating advanced intelligence into physical and digital workflows.
Email XERV Technical Team for Inquiries