Researching what comes next.
TAU CORE is Vecorps' internal research line for efficient personal AI. Before and alongside the development of TAU, we explored several model scales and architectures to better understand what personal AI could look like when computation, memory and adaptation move closer to the user.
That work continues to inform how we think about future TAU runtimes. TAU today runs Qwen3-4B. TAU CORE remains internal research.
What we are exploring
On-device efficiency
How capable personal AI can run within the memory, power and latency constraints of personal devices.
Adaptation
How personal AI can become more useful over time without requiring a central cloud profile.
Memory
How AI can preserve useful personal continuity while leaving memory visible and under user control.
Long-term personal intelligence
How models, memory and tools may eventually work together to create increasingly capable personal AI.
TAU CORE
Experimental edge model
TAU CORE 0.12B
An early compact research program exploring continual adaptation and highly constrained on-device inference.
Personal AI research model
TAU CORE 2.4B
A larger experimental program exploring hybrid sequence modelling, persistent personal context and local adaptation.
Future runtime research
TAU CORE 3B
A higher capacity research program exploring what a more capable future on-device TAU runtime could look like.
Some implementation details, training methods, evaluations and model artifacts remain proprietary while the work continues.