shahir.ai
Small utilities for people who run models on hardware they own. They open in the browser. Text you paste is not uploaded.
Flagship
Feel decode speed instead of reading a spec sheet. The same answer streams at a rate you set, from a constrained CPU to an extreme hosted pace, with a simulated wait for the first token.
More tools for private inference will be added here.
SAI is private on-premises AI. The model runs on computers in your building. Speed is then a property of that machine: how fast it reads a long prompt, and how fast it writes the reply. A hosted demo can hide both numbers. These tools exist so a team can see them.
The token speed visualizer is the first one. It does not call a model. It releases text at a tokens-per-second rate you choose, and it can hold the first token until a context size would have been read. Use it before a visit, or beside a machine you already run, to judge whether 10, 30, or 200 tokens per second is acceptable for notes, code, or an agent.
Nothing in this section replaces an install. SAI still goes on hardware you own, with on-site support, and it can run with outbound internet turned off. Read why SAI stays private and how it stays on site, or request an on-site visit.
Developed by shahir.ai.
Small in-browser utilities for people looking at private and local AI. The first is a token speed visualizer. More tools will be added on this page. They do not replace an on-premises SAI install.
No. The token speed visualizer tokenizes and streams entirely in your browser. Custom text is not sent to shahir.ai and is not included in the shareable link.
They are developed by shahir.ai, the public site for SAI — private on-premises AI that runs in your building.
