tropos

Route AI models by their track record, not their word.

A calibration-robust router for multi-model AI.

You have several models, cheap to expensive. Tropos corrects each model's self-reported confidence by an observable track record, then sends each task to the cheapest model that is actually reliable and escalates the rest. The longer it runs, the cheaper and more reliable it gets.

$ pip install tropos
start here

Walkthrough

The whole loop step by step, in real code with the real terminal output: define a fleet, route, report outcomes, watch the routing improve.

open the tour →
the evidence

Research dashboard

The hypotheses, figures, and results behind the method, including the real-agent pilots and the cost analysis.

view results →
the paper

Read the paper

The full write-up of the framework, the experiments, and the honest limits, as a PDF.

open PDF →
the code

Source on GitHub

Open source, Apache-2.0. Install it as a Python library or run it as an ACP agent inside your editor.

browse the repo →
Honest about what it does. You supply each model's confidence and the outcome of each run; Tropos does the correcting, routing, and learning. It does not call your models or grade results for you, and it helps most on a fleet of cheap, often overconfident models. Where the cheap model is already reliable, it is a no-op with no regression, a safety net rather than a generic booster.
Open source, Apache-2.0. Built by Teo Qing Cong Eugene. Code and paper on GitHub.