About

IWhy I built it

I have worked over the last decade in several roles in pricing, consulting and financial services. One of the key issues in all my roles was how much the tech stack became a bottleneck, because all SaaS was built around general workflows with a fixed UI. It would never work fully.

For me, being able to vibecode a minimal, simple surface for financial analysis, one that could handle more data than Excel but was not bound to a UI built by someone else, was an incredible breakthrough. On top of this, if I could get AI to understand my operating context and preferred way of doing things, I could spare myself the drudgery of setting up spreadsheets for hours before I could get any insights out of the data. In any case, the real work begins after you have found out with certainty what is going on in the underlying business, as the execution effort begins.

I started out by building the pricing plugin that is available on the critias-ai GitHub. I still felt that 1) I couldn't fully understand how the agent, running through a heavier and more complex harness like Claude Code or Codex, was reaching its conclusions; 2) it was still a UI built by someone else with a very general use case in mind; 3) dependency on these providers was discomforting; and 4) pure, unadulterated curiosity.

With the pricing plugin doing a lot of my work reasonably well, I decided to expand on it. What if I could build my own pricing agent with the plugin as the brain, and a very simple open-source body? As coding harnesses have made it easier to write software, I believe we are in the early innings of an open-source software renaissance. This of course applies to coding harnesses too, as the success of Pi has shown.

So I put my pricing plugin together with components taken from Pi. What emerged is, in my view, simultaneously revolutionary and banal. I am using less and less Excel for my analysis, and thankfully I haven't had to build PowerPoint presentations by hand over the last few months.

I put this project out publicly because I wanted to get in touch with peers and users who would find this kind of thing useful. I made sure to cover the various industries that need pricing expertise but might struggle to operationalise it.

I would love to hear from anyone who comes across this. Our codebase is available for you to make your own. Best of luck.

Danial Kodvavi

IIWhy it exists

Pricing teams live in spreadsheet exports, and the tools that promise to analyse them either cost a six-figure licence or make up numbers. Critias is a third option: an analyst you run yourself, whose every figure you can trace to the script that produced it.

IIIWhy open source

You should be able to see exactly how an analysis reached its answer. The app is licensed under Apache-2.0 and the Pricing Plugin under MIT. The evaluation suite is in the repository, so you can check how well it works before you trust it.

IVBuilt on Pi

Critias is built on Pi, the open-source agent harness (earendil-works/pi). Pi gives Critias its model layer, agent loop and tool calling, so one app can work with many model providers. Critias adds the pricing skills, the data checks and the figure check on top. If you already work with Pi, Critias is an example of what a domain-specific agent built on it looks like.

VContact

Email hello@critias.ai, or find Critias on GitHub and LinkedIn.