QANDY
Your personal financial agent.
QANDY builds AI agents that do real financial work for people. They find where you overpay, compare better options and handle the switch, but only after you approve it.
You ask. You approve. QANDY SAVE executes.
QANDY SAVE
QANDY SAVE is a free desktop app that runs on your own computer. It reviews the bills and contracts you choose to share, spots where you're paying more than you need to, and brings you a clear recommendation. Nothing changes until you say yes. Every month it reports exactly what it saved, net of any switching costs.
- Find. It reviews your bills and contracts and spots where you overpay.
- Approve. It compares suitable alternatives in plain language. You decide, always.
- Execute. Once you approve, it helps complete the switch and logs the saving in your monthly report.
It launches first in the USA and Europe with electricity, insurance, broadband and mobile bills, with China to follow. A mobile app is coming soon.
Join the waitlist →
QANDY 2.0 27B
QANDY 2.0 27B is our own 27-billion-parameter financial model, and it powers QANDY SAVE.
- Runs locally. It processes your documents on your own computer, so sensitive analysis happens where your data lives.
- End-to-end encrypted. The QANDY service around the app is encrypted end to end. Anything shared with an outside service happens only with your explicit authorisation.
- Open source at launch. The code goes public at launch, so anyone can review how it works.
- Built to scale. Local compute means each new user adds very little cloud cost.
QandyEval
QandyEval is our benchmark for financial agents. It measures whether a model can do the financial work people actually need done, under the rules in force on the day.
- Five domains: personal tax, corporate tax and accounting, Save, trading, and insurance
- Nine jurisdictions: United States, European Union (at EU level and per member state), United Kingdom, Mainland China, Japan, South Korea, Hong Kong SAR, Taiwan and Canada
- Strict grading: right numbers and right official forms, field by field. Answer keys come from calculation engines audited against primary sources. An item only counts if the model gets it right in every run (pass^k).
- The law on the day: every item states the date it applies to. The hardest tier is built on 2026 rule changes and refreshed every quarter.
- The permission test: in Save, agents get written authority and are graded on staying inside it. Anything outside that authority must go back to the user as a question.
- Simulated, never real: portals, inboxes and drives are simulated. Nothing is ever submitted to a real tax authority or insurer, and no personal data enters the eval set.
Head-to-head results for QANDY 2.0 27B against Claude Opus 5.5 and GPT 6.1 Sol will be published on the QandyEval page once all three models have been run on the same frozen item set.
Models and datasets on this page
| Model |
Size |
What it is |
| Qandy 1.0 27B |
27B |
Our first 27B model, from Starburst, our trading research line. It is a reasoning model for crypto perpetual-futures entries. This is a recipe and evaluation release; the weights are not included. |
Datasets
| Dataset |
Rows |
What it is |
| Qandy 1.0 27B data |
12,462 |
The SFT and KTO training sets and the two evaluation sets behind Qandy 1.0 27B: crypto perpetual-futures snapshots paired with reasoned LONG, SHORT or NO TRADE decisions and their outcomes. |
Links