On-Device AI Model
A Local AI Model for Payments
Tapsilat-o1 runs offline on your own machine and handles payment and finance tasks, such as finding anomalies in transaction data.
user@tapsilat-os:~
$ tapsilat --serve --model=tapsilat-o1
starting tapsilat cli...
loading tapsilat-o1 (4b params, 4-bit)... [done]
Model Architecture
A mixture-of-experts (MoE) model with 4B parameters, quantized to 4-bit so it answers quickly on modest hardware.
What It Can Do
Step-by-step reasoning and function calling, and it plugs into common evaluation harnesses.
System Requirements
8GB of unified memory is enough for full-context inference on your own machine. It works offline.