Reads what you already have
PDF and Word files are parsed directly. Scans and photos go through Tesseract OCR. Files up to 50 MB are checked by their actual bytes, not the extension they claim.
Document extraction · Private preview
Peel the data
off the page.
Upload an invoice, contract, receipt or scan. Docpeel works out what it is, pulls out typed fields with a confidence score for each, and flags the values worth a second look before you export them.
- PDF · DOCX · scans
- JSON · CSV · Excel
- Your choice of model
Header
invoice number
0412
98%
vendor name
Northline Studio
96%
Totals
total amount
720.00 GBP
94%
due date Review
03 Apr 2026
64%
How it works
One pass from file to fields you can trust.
Uploads are processed in the background while the page keeps you posted. When a value looks wrong, re-process the document with a different field list without uploading it again.
- 01 Upload PDF, DOCX, JPEG, PNG, TIFF
- 02 Read Text layer or OCR
- 03 Classify Six document types
- 04 Extract Template or your fields
- 05 Validate Typed checks, confidence
- 06 Review Then export
Features
Built for the work between upload and spreadsheet.
Knows what it is looking at
Each document is classified as an invoice, contract, receipt, tax form, legal filing or other, and extracted with a template built for that type.
Or asks for exactly your fields
Name the fields you need and their types at upload. Your list replaces the template and is remembered when you re-process the document.
Shows how sure it is
Every value carries a confidence score and is validated as a date, amount, email, phone, tax ID or JSON. Weak or invalid values are flagged for review.
Exports where you work
Download the fields as JSON, CSV or Excel, copy a single value, or download the original file you uploaded.
Handles a stack at once
Drop in up to 20 documents. Each one gets its own job and status, and the library lets you search by name and filter by type or status.
Document types
Field validation
Model providers
Your documents
Only the owner sees a document.
Every document, file and field is scoped to the account that uploaded it, twice: by database row-level security and again in the app’s own queries. The original is kept so you can check a value against it, and deleting a document removes the stored file too.
Text is sent to the model provider you configure for classification and extraction. Choose the provider whose data terms suit your documents.
Get Docpeel
In private preview.
There is no public sign-up yet, and the source is shared by invitation. If you have access, you can run your own instance against your own Supabase project and model key.
git clone git@github.com:aylith-labs/docpeel.git
cd docpeel/app && npm install
# Add your Supabase URL, anon key and model key
cp .env.example .env
# Create the tables, policies and storage bucket
supabase link --project-ref <your-project-ref>
supabase db push
npm run dev # http://localhost:5540
Needs repository access. Commands run in your own terminal.
Node.js 22 or newer
The app is a SvelteKit server. Run it on a long-lived Node host if you rely on background processing.
A Supabase project
The migrations create the tables, row-level security policies and a private storage bucket.
A model provider key
Anthropic by default; OpenAI, Google and OpenRouter work too. Your provider account and usage costs apply.
Not built yet
Highlighting the exact spot on the page a value came from, extraction history, a durable job queue, a public API, team accounts and billing. Throughput and accuracy have not been benchmarked.
Follow progress and the changelog