Faxes in. Source-linked data out.
Pepper reads the referrals, lab reports and notes your practice receives, pulls out the patients, results, medications and requests, and links each one to the printed line it came from. Staff check the result against the page instead of retyping it.
No payment card needed for the free trial. An API key comes with your account. During this preview, new accounts are by invitation; the examples, pricing and API guide are open to everyone.
Loading the recorded result…
Less retyping, and the source behind each finding
Less retyping
Patients, results, medications and requests come out as structured fields. Download them as a spreadsheet or send them to your own system through the API.
The source behind each finding
Each finding names the page and the printed line it came from. One click shows that line highlighted on the original, so anyone can check it.
Flags instead of guesses
A smudged digit, a page with no patient name, two patients in one fax: Pepper is built to flag these for a person rather than guess. The examples show it doing so, and where it falls short.
Three steps, no setup
Send a document
Upload a PDF in the workspace, or post it to the API. Long faxes are processed in the background and you can watch the progress.
Pepper reads and cross-checks
Typed pages are read from the PDF's own text. Scanned pages are read by one vision model; a separate local reader then checks the patient name, date of birth, dates and medication names against it, and what it reads differently is marked for a person. The numbers in a finding must appear on the line it cites, as that page was read. Premium adds a second read of the same page text by a stronger model and marks where the two differ.
Review, correct, export
See findings beside the original pages. Accept, correct or hold anything. Your edits are saved as a new version and the original reading is kept.
Fictional faxes, real results
Every example is a made-up document processed by the actual service. Open one and click through the findings.
Nephrology referral letter
A three-page referral from a primary care office: reason for referral, insurance, referring provider, medications, family history and recent results.
3 pagesdigitalnarrative
Scanned laboratory report
A two-page laboratory report scanned slightly crooked, with a prior-results column and a second page that carries no patient header.
2 pagesscanskew
Office visit note
A two-page visit note with symptoms the patient denies, vital signs, family and social history, medication changes and a late addendum.
2 pagesdigitalnegatives
Medication change and refill request
A one-page pharmacy fax, faint and sideways: one medication stopped, one started, one dose changed and one refill requested.
1 pagescanlow contrast
Mixed eight-page fax packet
An eight-page fax for two relatives with the same surname: cover sheet, laboratory panels, a referral note, a reprinted result, a shared clinic log, an unlabeled page and a medication list.
8 pagesfaxtwo patients
Clean one-page laboratory report
A clear, digital, one-page laboratory report with ten routine results.
1 pagedigitalcleanPepper beside a general AI model
A strong general model reads these documents about as well as Pepper does, and on some it does better. What Pepper adds surrounds the reading: the printed line behind each finding, fixed checks on names, dates and numbers, a review screen, and a much lower processing cost per page.
Held-out evaluation documents · 10 fictional documents · Not used to adjust the engine version that produced these figures. One run per system.
| Same files, same task | Keyed facts found exactly | Serious errors | Model usage per page | Average time per document |
|---|---|---|---|---|
| Pepper Standard gpt-6-luna, gpt-6.1-terra |
223 of 234 | 4 | $0.0009 | 25 s |
| Pepper Premium gpt-6-luna, gpt-6.1-terra |
222 of 234 | 4 | $0.0073 | 39 s |
| A general AI model, one request gpt-6.1-sol |
222 of 234 | 1 | $0.0177 | 36 s |
The six example documents · 6 fictional documents · Pepper's rules were adjusted on these six, so read them as a demonstration, not a test.
| Same files, same task | Keyed facts found exactly | Serious errors | Model usage per page | Average time per document |
|---|---|---|---|---|
| Pepper Standard gpt-6-luna, gpt-6.1-terra |
107 of 107 | 0 | $0.0007 | 24 s |
| Pepper Premium gpt-6-luna, gpt-6.1-terra |
107 of 107 | 0 | $0.0073 | 39 s |
| A general AI model, one request gpt-6.1-sol |
102 of 107 | 1 | $0.0215 | 40 s |
Counts on fictional documents from one generator, first run each, reusing none of the system's own saved work. "Serious errors" are a fact given to the wrong patient, a wrong value, date or present-or-absent marking, a value that is not printed where cited, an unreadable value reported as readable, or a required flag not raised. All 4 of Pepper Standard's serious errors on the held-out documents are one pattern: a stopped medication recorded as absent instead of as present and stopped. A manual check of those items against the pages found the drug, dose and action read correctly in each; the marking was one the output rules did not define when these runs were made, and the engine that made them was not consistent about it. The same check found that on the hardest scan Pepper read one drug name with one wrong letter (counted here as a missed fact) and Premium showed it as supported. Model usage is what the provider billed at list rates. The runs were made one after another, so much of each prompt was billed at the provider's lower rate for input it had just seen. With every such token at the ordinary input rate, the held-out figures are $0.00095, $0.00847 and $0.01833 per page. It is not an invoice and is not the customer price. These figures describe these documents; they are not a claim about real faxes.
Which engine. These runs were recorded with engine 0.2.0 and have not been repeated. The engine in service is 0.2.3. It was changed after these runs, partly because of what these documents showed: a medication reported as stopped is now recorded as present with the action stopped, a medication name the second reader spells differently is flagged, a contradiction is carried on the findings it concerns, and the second read no longer counts wording as a disagreement. For the engine in service these documents are regression cases, not untouched test documents, and the figures above do not measure it.
Standard for everyday faxes, Premium for the hard ones
You choose per document and see the exact price before anything is processed.
Standard
Proposed priceEveryday faxes: referrals, labs, visit notes, refills.
- Reads typed and scanned pages, including rotated ones
- Patients, results, medications, requests and a short summary
- The page and printed line behind each finding
- Flags what a person should check
- Up to 30 pages per document
About 50 pages fit in the free $1.00.
Premium
Proposed priceLong or difficult packets: a second read by a stronger model and cross-page checks.
- Everything in Standard
- A second read of the document by a stronger model; where the two reads differ on a date, a unit, a number or whether something is present, the finding is marked for review
- A check across pages for statements that contradict each other, on documents longer than four pages
- Referral intake fields: insurance, referring provider, contact details
- Up to 150 pages per document
About 12 pages fit in the free $1.00.
What the test documents showed. On 10 held-out fictional documents, one run each, recorded with engine 0.2.0, Standard found 223 of 234 keyed facts exactly and Premium 222 of 234: Premium did not find more facts. What Premium added was checking. Counted with the comparison the engine uses today, its second read disagreed with the first about a date, a unit, a number or whether something is present in 30 findings and recorded 23 more, and all of those are marked for a person to review (66 findings marked for review, against 22 on Standard). Its summaries covered more of the facts that matter (88% against 77%), and it took longer (about 39 s a document against 25 s). Ten documents are too few to say which tier is more accurate. The engine in service is 0.2.3; these runs were not repeated with it, and these documents helped shape its changes.
Premium is a deeper automated check, not a person at Tathya reviewing your document, and neither tier promises a perfect result. Both keep each account's documents separate, show their evidence and mark what is uncertain. Pricing details and a cost estimator.
One API under the same workspace
- An API key the moment your account is created
- Post a PDF, get a job, read the result: the same contract the workspace uses
- Safe to retry: a repeated request never charges twice
- Predictable errors that say what to do next
curl -X POST "https://parsedocs.tathya.ai/v1/parse?tier=standard&data_class=deidentified" \
-H "Authorization: Bearer $PARSEDOCS_KEY" \
-F "file=@fax.pdf"
{
"job": {
"id": "job_3fK2…",
"status": "queued",
"pages": 3,
"estimate": { "amount": "$0.06" }
},
"credit": { "available": "$0.94" }
}Start with $1.00, no card
- Create an account with your email. You get $1.00 of usage credit, once.
- Use the built-in examples, or a file that had patient information removed before upload.
- You see the price before each document and the balance after it.
- When the credit runs out, your document is kept. Continue with a payment and a business associate agreement; once both are complete and we have reviewed your practice details, it starts by itself.
About patient documents
Real patient documents need a signed business associate agreement with Tathya first, whatever your credit balance. Until then the service refuses them before the file is even received.
Pepper does not diagnose, grade urgency or decide care. It reads what is printed and shows where.