Transparency
How {brand} works, and where it doesn't
Most AI tools ask you to trust an answer. Quarria is built so you don't have to take anything on faith. This page explains how results are produced, what happens to your data, and what Quarria will not do.
The method
How a signal is produced
- 1
Scan
On each run, Quarria’s agents search public sources relevant to the question you defined.
- 2
Cite
Every claim in a signal carries its source: title, publisher, date, and a working link. If it cannot be sourced, it is not shown.
- 3
Rank
Signals are scored for relevance to your context, built from what you have told Quarria about your world and what you have engaged with, not a generic average.
- 4
Act
Each signal ends in a recommended next move, so intelligence becomes motion rather than backlog.
The AI itself
An honest account of the model
Which models
Quarria runs on large language models, currently Anthropic’s Claude family. You choose the depth of analysis per project, and the cost in credits is shown before you spend it.
AI can be wrong
Language models misread, overweight, and occasionally misstate. That is exactly why every claim is cited: check the source before acting on anything that matters. Output is advisory. It informs your judgment, it does not replace it.
Unsourced claims are dropped
The pipeline enforces citation and discards claims it cannot attribute rather than guessing. A signal without a working source does not ship.
The public feed is reviewed
Issues of the public signal feed are generated by Quarria and read by a person before publication. They carry the same citation standard and say so on the page.
Your data
What happens to what you put in
Your context is yours
Your profile, projects, notes, and feedback exist for one purpose: making your results more relevant. That compounding works for you and only for you.
We do not train on your data
Quarria never uses your content or context to train or fine-tune any model, and never sells it. Your requests are processed by our model providers to answer them; a signed data-processing agreement with those providers is on our roadmap and not in place yet, and we would rather say so than imply otherwise.
Accounts are isolated
Each account's context, signals, and learning are stored against that account and read back only for it. One customer's data never informs another customer's results.
What we record about the world
Quarria keeps counts about public sources and public stories: which public sites and which public articles have appeared in results. The counts carry no account, project, note, context or text of any customer, so they cannot be traced back to anyone, and today they are only recorded, not used to change anyone's results. If that changes, this page will say so first.
You stay in control
You can view, edit and export your context at any time, and delete your account and its data yourself from your Profile.
The full picture of controls, and every third party involved: Trust and security
The boundary
What {brand} is not
- Not legal, financial, medical, or investment advice. Signals are information with a suggested move, and consequential decisions deserve professional judgment.
- Not autonomous. Quarria recommends, you decide. It takes no external action on your behalf.
- Not surveillance. It reads public sources and the accounts you explicitly connect. Nothing else.
Something we should hold ourselves to?
If a claim here is unclear, or there is a standard you expect and we have not listed, tell us.