Finance and AP teams running ERPNext still key most supplier documents by hand. A PDF invoice arrives by email, someone opens it, retypes the supplier, the dates, every line item, the tax, and the totals into a new Purchase Invoice, then files the original in a shared drive. ERPNext core gives you the destination DocTypes and a clean posting workflow, but nothing in the box reads an inbound document — there is no OCR, no email-to-invoice ingestion, and no AI that maps unstructured text to Purchase Invoice Item rows. As volume grows this manual lane becomes the bottleneck: slow month-end closes, transposition errors, missed early-payment discounts, and no audit trail linking the posted record back to the source file.

ECOSIRE builds a dedicated Frappe app that closes that gap end to end. We ship a proper custom app (its own module, DocTypes, and hooks.py) rather than a pile of loose customizations. Inbound documents arrive two ways: a monitored email inbox (via ERPNext Email Account incoming rules and a scheduler job that pulls attachments) and a drag-and-drop upload screen backed by a File-attached staging DocType we call Document Extraction Job. Each job runs the PDF or image through an AI/OCR pipeline — a configurable extraction provider plus a normalization layer — and the parsed result is written to structured child-table fields: header data (supplier, invoice number, dates, currency, totals) and a line-item table with description, quantity, rate, and tax. A confidence score is stored per field so low-confidence extractions route to human review instead of posting silently.

Technically the automation lives where Frappe expects it. A before_insert/on_update server-side flow (registered through doc_events in hooks.py) validates the extracted payload, runs supplier auto-matching against the Supplier master and item auto-matching against Item (fuzzy match on name, supplier part number, or a learned alias table), applies your tax templates, and creates a draft Purchase Invoice or Expense Claim linked back to the source File. Whitelisted methods expose the pipeline to the REST API so an external mail gateway or RPA bot can post documents in, and client scripts add a review-and-approve UI on the job form. Scheduler events (scheduler_events in hooks.py) poll the inbox and retry failed jobs. Access is controlled with a dedicated Role Profile and DocType permissions so AP clerks can review while only approvers can submit the resulting invoice. It targets Frappe/ERPNext v15 and v16.

Because this is build-to-order, nothing is a generic download. We start with a short scoping call, confirm your document types, tax rules, matching logic, and target DocTypes in writing, then build against that scope — typical delivery is 2 to 4 weeks from confirmed scope depending on the number of document layouts and integrations. You get the installable source, a UAT pass on a staging bench with a rollback plan, documentation, a training session, and a post-go-live support window. ECOSIRE installs it on your bench (self-hosted or Frappe Cloud private bench) and hands over the git repository so the app is yours to run and extend.