Snowpine builds four things for mental health practices: the letters, the superbills, a private clinical model built to take a patient's real name once your BAA is signed, and the visibility work that decides whether a searching patient ever finds you. Today it runs in demo mode, on fictional patients only; no patient data enters until the BAA path is live.
Each product drafts the paperwork for a job that currently lands on you personally; you review and sign, always. Three of the four are built to run inside a boundary your practice can put a patient's name into, once your BAA is signed.
IEP and 504 support, work and school excuse notes, and prior authorization appeals, drafted from the letters you have already signed and written in your own phrasing. Every draft tells you which prior letters it drew from so you can check its reasoning; the proof below shows two of those pairings, sentence against sentence.
Built from your schedule or from a photographed paper day sheet, then checked against the nine reasons claims actually come back before your patient ever sees one. Where it is not certain it flags the field, and it never fills a gap by guessing. Not yet proven on real handwriting.
A clinical assistant under a signed BAA that takes the patient's real name and real numbers, and answers the questions consumer chatbots refuse to touch. It cites, or it says it does not know.
Get found by patients searching Google and by patients asking ChatGPT, using nothing your licensing board would object to. No testimonials, no outcome claims, no PHI, anywhere.
See the deliverables, honestly labeled: a sample practice page built to be citable, and a sample monthly citation report. Both fictional; we publish a real practice's numbers only when a real practice says yes.
Everything below is fictional and labeled as such. We gave the system three of a fictional psychiatrist's past signed letters, then asked for a brand-new IEP letter in one line: "K.M., DOB 9/14/2016, Roseville Joint Union School District, generalized anxiety disorder, sertraline 50 mg daily, seen every six weeks since January 2026, freezes on timed tests, Monday absences around transitions, CBT with outside therapist since February. Due Thursday." This came back. Below is the letter body word for word, no text changed, with one addition: the clause the reviewing clinician would strike is shown struck. The full output adds her letterhead and a bracketed deadline slot above the salutation.
The struck clause is the causal claim the engine added without a source; striking it is the signer's job the page keeps talking about, shown rather than described. One more true wrinkle: the request typed the district's name incompletely (its legal name is Roseville Joint Union High School District) and the draft carried the typo faithfully, because it drafts from what you give it.
To the IEP Team, Roseville Joint Union School District
Re: K.M., DOB 9/14/2016
I am the treating psychiatrist for K.M., whom I have followed every six weeks since January 2026 for generalized anxiety disorder. K.M. is currently prescribed sertraline 50 mg daily. K.M. has also been participating in cognitive behavioral therapy with an outside therapist since February.
This condition adversely affects K.M.'s educational performance in specific, observable ways. Under time pressure, K.M. freezes on timed tests and is not able to demonstrate knowledge accurately or consistently. In addition, K.M. has a pattern of Monday absences around transitions, which is consistent with anxiety-related difficulty anticipating and managing changes in routine and re-entry to school demands.
I am requesting the following accommodations, each tied to the impairment above: extended time on timed assessments, because K.M.'s performance is significantly affected by time pressure rather than lack of knowledge; testing in a reduced-distraction, low-pressure environment, because anxiety escalates in more stimulating group settings; and support around Monday and other transition periods, because these are predictable times of increased anxiety and school avoidance.
I expect these supports to be needed for at least the current academic year, with reassessment based on clinical progress. I am available to the team for questions through my office.
Sincerely,
Maya Chen, MD (fictional)
Her signed letter (fictional, ADHD, 2025): "I am the treating psychiatrist for A.R., whom I have followed every six weeks since September 2025 for attention-deficit/hyperactivity disorder, combined presentation."
The new draft: "I am the treating psychiatrist for K.M., whom I have followed every six weeks since January 2026 for generalized anxiety disorder."
Her signed letter: "extended time (time and a half) on all timed assessments, testing in a reduced-distraction environment, and scheduled movement breaks every 30 minutes."
The new draft: "extended time on timed assessments, because K.M.'s performance is significantly affected by time pressure rather than lack of knowledge; testing in a reduced-distraction, low-pressure environment; and support around Monday and other transition periods."
Note what changed: the ADHD movement breaks did not carry over to an anxiety letter, and transition support appeared because the request mentioned Monday absences. The structure and voice are hers; the clinical content follows the new facts.
What the draft added or dropped, called out honestly: it wrote "anxiety escalates in more stimulating group settings," a causal claim that was not in the request; the clinician strikes or owns that line. It dropped "time and a half" from the source letter's extended-time accommodation; the number a school can implement has to be restored by the signer. And "support around Monday and other transition periods" is not yet an implementable service; the IEP team needs a named support, staff role, and duration. Every draft ends at your signature for exactly these reasons. License number, date of exam, and observed-versus-reported statements are template fields the engine leaves empty for you. Two more things a careful reader will catch: the draft names the medication and dose, and a dose enters a school file only with the parent's consent to disclose, a confirmation that belongs to the signer, because nothing leaves without the clinician sending it; and because it requests accommodations without specialized instruction, this letter functions as 504 support unless the clinician adds an instruction request, which is the clinician's call.
You are not evaluating a chatbot. You are deciding whether to put a patient's identity into someone else's system.
Yes, once the BAA is signed. You do not have to de-identify, invent initials, or write around the chart.
No. There is no setting to find and no default to reverse, because the pathway does not exist.
Treat it like the chart. Retention follows your practice's schedule: your regulator sets the floor, which for California mental health records, and for minors especially, is measured in years, not a setting. We delete on your instruction above that floor, and we will refuse a deletion that would put you below it. Our security page states our current posture plainly; the full packet your attorney will actually want arrives with the BAA.
No routine human review. Support access requires your written, time-limited approval, and every instance lands in an audit log you can pull yourself.
It is built to take the questions consumer tools refuse rather than deflect them, and it will not tell a board-certified psychiatrist to consult a healthcare professional. Judge it on the one full exchange we publish below, not on this sentence.
No. There is no recording or ambient-capture feature, and none is turned on quietly later; if we ever build one, it ships only with the consent flow your state requires and is announced as a change to this page.
The question, exactly as typed into the demo desk. Fictional patient, labeled as such:
"34-year-old with bipolar I, stable three years on lithium 900 mg nightly (last level 0.8), just found out she is 6 weeks pregnant and wants to keep the pregnancy. Continue, switch, or taper? What monitoring schedule?"
A consumer chatbot refuses this or tells a psychiatrist to see a doctor. Below is how the desk actually answered, condensed from the full response.
"For a patient euthymic three years on lithium, abrupt discontinuation is usually the worst option. Continuation with informed consent and intensified monitoring is often a reasonable, evidence-supported choice… lithium trough plus renal function every 3 to 4 weeks until 34 weeks, then weekly until delivery; thyroid each trimester; anatomy scan and fetal echo around 18 to 22 weeks after first-trimester exposure."
Sources it named: FDA lithium labeling; Patorno et al., NEJM 2017; Viguera et al. on discontinuation relapse risk; the ACOG practice guideline; NICE antenatal guidance.
It also said what the evidence does not settle: no randomized trials answer continue versus stop, and switching an already-stable responder has limited comparative evidence. It cites, or it says it does not know.
Read the full answer, unedited, on its own page. The complete output names the first-trimester cardiac risk, the 18 to 22 week fetal echo window, peripartum level management, and coordination with maternal-fetal medicine.
A tool that claims no limits is not a tool you should put a patient into. These are ours, and every one of them is deliberate.
AI tools for mental health practices. It drafts IEP and 504 letters, excuse notes, and prior authorization appeals, generates checked superbills, and makes a practice findable to patients searching Google or asking an AI assistant. The private clinical model will run under a Business Associate Agreement; until that path is live it takes no patient identifiers.
No, not today, because today it refuses the data HIPAA covers: Snowpine runs in demo mode and does not accept protected health information. Before any workspace takes patient data, we sign a BAA with the practice, on every plan, at no extra cost, with the safeguards HIPAA requires specified control by control in the BAA packet rather than adjectives on this page. Your practice keeps its own policies and training.
You are. Snowpine produces a draft, never a signed clinical document. Nothing is sent, filed, or submitted automatically, and the signature is always yours.
Yes, with one honest caveat. One photograph of your day sheet produces the day's superbills for every clean row, and rows the checker flags come back to you as questions before anything prints; answering those is the only data entry left. Not yet proven on real handwriting.
No. Snowpine sits beside them and handles the documents they handle badly. Superbills export as PDF and CSV, and we never submit claims.
The board risks in physician advertising are testimonials, outcome claims, solicitation of people in treatment, and anything deceptive. We build none of those: no patient testimonials, no review campaigns aimed at people in treatment, no outcome claims, no protected health information. This is our design constraint, not legal advice; your board's rules are the ones that bind you.
Practicing psychiatrists, psychologists, therapists, and psychiatric nurse practitioners in solo practice or small groups, especially out-of-network practices where the letter volume lands on the clinician.
Snowpine is in demo mode today: no protected health information enters the system until your BAA is signed and the private hosting path is live. What starts tonight is the work that never touches patient data, the visibility work and the clinical desk with no patient identifiers. Letters and superbills turn on the day your BAA is signed, at no extra cost. Ten practices, then the door closes at this price.
Cancel any month. If the BAA does not hold up under your review, first month refunded, no questions.
We send the agreement, the subprocessor list, and the security document first.
Set up a workspace only if they hold up. Read the security page now, no email needed.
Used only to send the packet and reply. Privacy.