# Critias — open-source pricing agent that shows its working

Ask why gross margin moved and get an answer where every figure comes from a script that ran on your data. Open source, Apache-2.0, runs on macOS and Windows.

Critias is an open-source pricing agent. Attach a sales export, ask why gross margin moved, and get an answer in which every figure comes from a script that ran on your data.

## Every figure has a source.

Critias writes Python, runs it on your files and reads the output. Before a reply reaches you, every figure in it is matched against what a script printed. When a number cannot be calculated from the data, Critias says so and leaves it out. It does not estimate.

[How it works](https://critias.ai/how-it-works)

## One question, start to finish.

A sped-up recording of Critias answering a margin question on sample data: the file, the questions it asks, the scripts it runs and the checked chart.

[Watch the demo](https://critias.ai/demo)

## No data is sent to Critias.

Critias runs on your computer and has no server of its own. Files are stored and computed locally, in DuckDB and Python. The model you choose sees your question, the context you confirmed, column profiles, rows the data checks flagged, and the output of the scripts it runs. Choose a provider whose data terms you accept.

[Where your data goes](https://critias.ai/security)

## Seven skills, in Critias or your own agent.

Margin and price-volume-mix analysis, opportunity scans, recommendations, price structure design and rollout planning, for manufacturers, distributors, ecommerce, SaaS and subscription businesses. Use them inside Critias, or add the open-source Pricing Plugin to the AI agent you already use.

- Intake
- Analyse
- Scan
- Recommend
- Design
- Execute
- Triage

[The Pricing Plugin](https://critias.ai/plugin)

## Results you can rerun.

The evaluation suite ships with the code: synthetic and public datasets, answers computed by scripts, and a pass only when every figure matches. Failures are published alongside the passes.

[Evaluations](https://critias.ai/evaluations)

## Start with one question.

```bash
git clone https://github.com/critias-ai/critias-pricing-agent.git
```

Open source. You need Node.js, Python and a key for the model provider you choose.

[Full install guide](https://critias.ai/install)