Questions & answers

Aurora reads a record of messages between two people and returns a written brief that names the recurring patterns it finds — things like schedule control, contradictory accounts, isolation, demeaning, financial pressure — and links each one to the exact messages behind it. Coercive control rarely lives in any single message; it shows up as a pattern across hundreds of them, which is hard to see by hand. Aurora reads for that pattern.

On purpose. An account means an email, a password, and a standing record that you signed up — the wrong thing to ask when your inbox or your devices aren't fully yours to control. When you start a run, you get an access code instead — what your payment attaches to, and how you get back in — plus a passphrase that locks your finished file. Save both somewhere only you can find them, like a password manager or a private note; guard the passphrase especially, since we don't keep a copy of it, and if it's lost, even we can't reopen your file for you.

People preparing for a family-law proceeding, and the people who help them — a pro-se litigant getting ready for a hearing, an attorney building a record, an advocate or evaluator who needs to read a long history quickly. It's a preparation tool, used in the weeks or months before a proceeding. It is not a crisis or safety tool, and it doesn't do anything in real time.

Two ways to run the same analysis. Aurora Cloud is the hosted version: you upload a message export and Aurora runs it on our servers, then deletes it. You pay per analysis, and it's open to anyone preparing for a proceeding. Aurora Bench runs on your own computer, using your own AI account — so the case never leaves your machine and never touches our servers. It's built for attorneys, GALs, evaluators, and advocates who handle many matters, and it's in private beta right now; if that's you, ask us for access.

No. Aurora names patterns and cites messages. It does not reach legal conclusions, decide your case, or tell you what to do — and it is not a substitute for a lawyer. Review every citation yourself, and talk to a qualified attorney about your situation.

No. It does not diagnose people or label anyone as an abuser. It describes what the messages do, over time, and shows you the messages. The reading and the judgment stay with you, your attorney, and the court.

No, in two senses. It doesn't pick and choose which messages to read — it reads the whole record. And it doesn't decide what matters for your case or what belongs in front of a court — you and your attorney do. Aurora's job is to surface what's already in the record so a person can decide what to do with it.

No — and this is an important boundary. Chain of custody is about how your messages were collected and preserved before they reach Aurora: that's the job of your export tool or a forensic examiner, not Aurora. Aurora analyzes the export you give it; it does not certify where that export came from or that it is complete and unaltered.

What Aurora does record is its own work — which version of its method and which version of the AI produced your brief — so the analysis itself is reproducible and attributable. Think of it as documenting how the reading was done, not vouching for how the messages were gathered.

Your messages are the evidence. Aurora's brief is a reading aid that helps you and your attorney find and organize what's already there — closer to attorney work product than to an exhibit. How anything gets used in a proceeding is a decision for your attorney and the court.

Neither, by design. Aurora will not invent patterns that aren't in your messages — every claim is tied to specific messages you can check. And it is not guaranteed to find every pattern present: a pattern Aurora doesn't surface is not proof that none exists. It's a careful reader, not an oracle.

You can, and for some people that's the right call — the underlying approach is just careful prompting of a capable AI. What Aurora adds is a maintained, versioned method built specifically for this: a pattern catalog grounded in the established literature, discipline that footnotes every claim to a real message and rejects ones it can't ground, and an output you can hand to an attorney. The honest short answer: you're paying for the method and the rigor, not for access to an AI.

The pattern catalog comes from the established academic literature on coercive control — Evan Stark, Lisa Aronson Fontes, and Emma Katz — not invented for the product. The method is versioned and cited in every brief, so a reader can check the source.

Aurora's goal is to stay within reach of the people who need it most — someone who can't afford a lawyer should still be able to walk into court prepared. That's why Aurora is pay-what-you-can: there's a standard per-analysis price, and if it's out of reach, you pay what you can.

You'll never be caught off guard: every charge is per analysis — no subscription — and before anything runs you're given a price on the pay-what-you-can spectrum. Nothing is charged that you haven't said yes to. If a paid analysis doesn't finish, or it doesn't serve you, ask within 7 days and we'll refund it — no form, no fight. We run this on good faith, and we just ask the same in return.

Courts limit how long a filing can be, and a raw message history can run to hundreds or thousands of pages. Aurora consolidates that sprawl into a focused, cited brief — the patterns that matter, each tied back to specific messages — so you and your attorney can build a filing that fits the limit instead of attaching the whole record. It doesn't write the filing for you; it gives you a far shorter, organized starting point.

Right now Aurora reads a PDF export from a co-parenting platform — OurFamilyWizard, Civil Communicator, 2houses, TalkingParents, or AppClose. We add support for new sources as people need them. If you're not sure whether your export will work, get in touch.

This is about Aurora Cloud, the hosted version. Your export is read in your browser first; to run the analysis your messages are sent to our AI provider and held on our servers only while it runs, then deleted the moment your brief is ready. The provider deletes them within 30 days and never uses them to train its AI — rarely, a message its safety systems flag may be kept a little longer. Your finished brief is an encrypted file only you can open; it stays downloadable for 7 days, then it's deleted too.

With Aurora Bench, none of this applies — the analysis runs on your own computer, so your messages never leave your machine.

The full account — how your data is handled →