Any document. Every field.
With its source.
dScribe reads the documents your team retypes today — invoices, packing lists, certificates, claims, onboarding forms — and returns every field with a confidence score and the exact place in the document it came from. It started in customs. It works anywhere a PDF has to become a record in a system.
5
Output formats: JSON, XML, CSV, fixed-width, EDIFACT
3
Delivery routes: REST, SFTP, webhook
Any
Document type, from a sample or a schema
The same engine, whatever the document
Every industry has the same job hiding in it: a document arrives, someone reads it, and they type what it says into a system. A freight forwarder does it with commercial invoices. An accounts team does it with supplier bills. An insurer does it with claim forms. A clinic does it with vaccination records. The document changes; the work does not.
dScribe is built around that fact. You describe the document once, as a schema, or hand it a sample and let it infer one. From then on every document of that type is read the same way: each field extracted, scored for confidence, and tied back to the position in the page it was read from. What it cannot find, it lists as missing. It never fills a gap with a plausible guess.
The output goes wherever your system expects it, in the shape your system expects. JSON to an API. XML or CSV to an import folder over SFTP. A fixed-width record to a mainframe. An EDIFACT message to a trading partner. The same extraction, five formats, no re-keying.
Where dScribe is used
Customs is where it started. These are the documents it reads today.
Customs & freight
Commercial invoices, packing lists, transport documents and certificates, cross-checked against each other and drafted into ENS and CDS declarations.
Accounting & finance
Supplier invoices and credit notes read into the ledger, with totals reconciled and the supplier resolved against your master data. Delivery to PennyLane or any ERP.
Manufacturing & distribution
Purchase orders, delivery notes, certificates of conformity and material test reports, extracted and dropped into the ERP as XML or fixed-width.
Insurance & legal
Claim forms, policy schedules and contracts: the parties, dates, amounts and clauses pulled out with the page reference for each.
Health & HR records
Vaccination and health certificates, onboarding forms, licences. A schema inferred from one sample record, then applied to the rest.
Your document
If it is a PDF, an image or an email attachment, dScribe can read it. Send us three examples and we will show you the fields the same day.
From document to record
Six stages, each one visible and auditable.
Describe the document
Pick one of the shipped schemas, write your own, or upload a sample and let dScribe infer the fields from it.
Documents arrive
Upload in the app, forward by email, drop on SFTP, or post to the API. Scanned or digital, PDF or image.
Extract with evidence
Every field gets a value, a confidence score and the place in the document it came from. Missing fields are listed as missing.
Check
Documents that should agree are compared: invoice totals against line items, packing-list weights against the invoice. Parties are resolved against your master data, and an ambiguous match is refused rather than guessed.
Review what is flagged
Your team sees only the fields that need a human. Confirm or correct in one screen; the correction is kept for next time.
Deliver
JSON, XML, CSV, fixed-width or EDIFACT, over REST, SFTP or webhook, or through a connector to SAP, Salesforce, Shopify, PennyLane or ELO.
Built to be trusted
The parts that matter when the output goes into a system of record.
Confidence on every field
High-confidence values pass. Anything uncertain is flagged. You set the threshold per field, so a reference number can be stricter than a description.
Provenance
Each value carries the page and position it was read from. An auditor, or your own team, can see why a number is what it is.
Gaps, not guesses
A field no document supplied is reported as missing. A draft quietly filled with plausible values is worse than an obviously incomplete one.
Schema inference
Hand dScribe one sample of a document it has never seen and it proposes the schema. Review the fields, save, and the next hundred are read the same way.
Master data resolution
Supplier names, addresses and codes are matched to your registers. One clear match resolves; two close matches ask.
Any output, any route
Templates for JSON, XML, CSV, fixed-width and EDIFACT, with the escaping and padding handled for you. Delivery over REST, SFTP (host-key pinned) or webhook.
Human review built in
Review is a step in the pipeline, not an afterthought. Nothing leaves without passing the rules you set.
Audit trail
Every extraction, correction and delivery is logged with who, when and what changed.
Private Cloud or shared
Run today as your own deployment, built around your documents and systems. A shared subscribe-and-go platform is coming.
Common questions
Any PDF or image with a repeatable structure: invoices, packing lists, transport documents, certificates, purchase orders, delivery notes, claim forms, contracts, health and HR records. Scanned documents go through built-in OCR first. If you have a document type we have not seen, send three samples and we will infer a schema from them.
We do not publish a single accuracy number, because it depends on the document and the field. Instead every field is scored, low-confidence values are flagged for a person, and we run a pilot on your own documents so you see the real rate before you commit.
No. dScribe feeds them. It turns a document into a record in the format your system already imports, and sends it there.
Today each client runs their own Private Cloud deployment of Droblet, so documents and extracted data stay in your tenant. A shared platform with per-tenant isolation is in development.
Yes. The Droblet MCP server exposes extraction, case checks and classification as tools inside any MCP client, with OAuth 2.1 and no ability to file anything without a human approval in Droblet.
Send us three documents
We will show you the fields, the confidence scores and the output in your format, the same day.