Jeremy Neely Text Jeremy

AI & Agents · Production

TokenX

Measures what each Hatz AI model really costs, then helps people choose.

Built for: AME staff and Secure AI clients choosing AI models

Scrolling the model ranking, opening the best-balance pick and filtering the full model table (measured sample).
The model guide asking what you are doing and what matters most.
The model guide asking what you are doing and what matters most.

The problem

Hatz offers dozens of AI models but does not publish a per-model credit rate, and credits post in delayed batches. Without real numbers, people overspend or pick a cheap model that fails at the job.

What I built

I built a Python command-line tool that runs the same chat on every Hatz model and watches the credit meter to measure cost per 1,000 tokens. The results feed a plain-language HTML model guide that also ships inside Secure AI for Excel.

How it works

  1. Take a baseline reading of the credit meter.
  2. Run a fixed sample chat on one model.
  3. Wait for the credits to post and compute the rate.
  4. Retest suspect results and save progress safely.
  5. Merge costs with public benchmarks into the HTML guide.
How TokenX is wiredStandard-library Python only. The API key is encrypted with Windows DPAPI and never written in clear text.
Comes inInside TokenXGoes out Hatz model list dozens of models Sample chat fixed, multi-turn Credit meter posts in batches Baseline reading credits before Run the chat tokens reported exactly Wait & compute credits per 1,000 tokens Retest suspects up to three times Results file grows as models appear Model guide HTML, plain language

Built with

  • Python 3.10+ (stdlib only)
  • unittest
  • Windows DPAPI
  • Hatz AI REST API
  • Node.js
  • HTML/CSS/JavaScript
Text Jeremy about this Next: FollowUp AI All 38 tools Texting opens your Messages app with a note already written. He replies himself.

More screens

The best match, runner-up and also-consider models with pros and cons.
The best match, runner-up and also-consider models with pros and cons.
Three models compared side by side on cost and skill.
Three models compared side by side on cost and skill.
A scatter plot of cost against Excel skill with a sweet-spot area.
A scatter plot of cost against Excel skill with a sweet-spot area.
Scrolling the model ranking, opening the best-balance pick and filtering the full model table (measured sample), a frame from the demo clip (sample data).
Scrolling the model ranking, opening the best-balance pick and filtering the full model table (measured sample), a frame from the demo clip (sample data).
Scrolling the model ranking, opening the best-balance pick and filtering the full model table (measured sample), a frame from the demo clip (sample data).
Scrolling the model ranking, opening the best-balance pick and filtering the full model table (measured sample), a frame from the demo clip (sample data).

More of Jeremy's work

All 38 tools, by category