Use “Capture a receipt” in Recued. Read one receipt locally — OCR for a photograph, text extraction for a PDF — pull out the merchant, date, total, VAT and payment method, and keep it as an expense.
Capture a receipt
Use “Capture a receipt” in Recued. Read one receipt locally — OCR for a photograph, text extraction for a PDF — pull out the merchant, date, total, VAT and payment method, and keep it as an expense.
Complete workflow
Expense Ledger
7 pinned recipes
Use Expense Ledger in Recued for expenses, receipts, bookkeeping, and ocr. It includes 7 built-in actions and 6 ready-to-run workflows. Actions that change data use Recued's approval controls.
You are viewing one recipe in this workflow. Installing it opens the complete pack so its other recipes and background automations arrive together.
Use “Review an expense” in Recued. Confirm or correct one captured expense. Whatever you leave blank keeps what was extracted; whatever you fill in replaces it.
Use “Show the expenses” in Recued. List captured expenses for a period with a running total, and count what still needs review. Read-only — it re-reads no receipt and re-runs no extraction.
Use “Export a period” in Recued. Produce the CSV for one month — every reviewed expense, with totals. Only rows you have reviewed are included, so the export is what you stand behind rather than what the model read.
Use “Import expenses (CSV)” in Recued. Reads a card or expense export you downloaded yourself, lets you match its columns from the file’s own header, and records every line as an expense. Re-importing an overlapping export adds only what is new.
Use “Records runtime” in Recued. It helps you complete this workflow using the connections and settings you choose.
Show all 7 recipes
Use “One expense” in Recued. Open one expense from the list that names it, and read everything recorded about that expense on one page.
How a receipt is read
A photograph (image/*) goes through tesseract OCR. A PDF goes through pdftotext. Both are small local tools; nothing is uploaded to a scanning service.
⛔⛔ A scanned PDF produces NO TEXT, and pdftotext reports that as SUCCESS. Verified live on 2026-07-30: a receipt image saved as a PDF came back exit 0 with zero characters. That is the ordinary case for anyone who scans rather than photographs, and it is the dangerous one — an empty string handed to the extractor returns a row of nulls that is indistinguishable from a real receipt with nothing on it.
⇒ This recipe checks the extracted text is non-empty before the model ever sees it. A PDF with no text layer is recorded no_text with a note telling you what to do — re-add it as a JPG or PNG and it will be OCR'd. It is never guessed at.
⚠ There is no local path from a scanned PDF to text. pdfimages can only LIST the images inside a PDF, not extract them, so the pack cannot pull the page out and OCR it for you. Photographing the receipt directly is the working path.
What is stored, and what is not
The extracted TEXT is kept as its own record so you can see what the model actually read. The receipt file itself is untouched — this recipe copies nothing and deletes nothing.
⚠ Every field is AI-extracted and unverified. The extractor is told not to invent a total, tax, date or merchant that is not printed, and to report its own confidence; a low-confidence row is still stored, marked needs_review. Nothing here is booked to your accounts — this is a ledger of what was read, which you confirm in Review an expense.
⚠ A likely duplicate is FLAGGED, never refused. If a receipt matching an existing merchant, date and total is captured again, the new row is kept and marked duplicate_suspect. Refusing it would be the wrong trade: photographing the same receipt twice is cheap to sort out, and a silently dropped receipt is an expense you never claim.
How it works
Inspect the data fetches, transforms, gates, and output this recipe runs.
Process (45 steps)
record_id
trim
Trim whitespace from setting record id
id_present
compare
Check if record id has data
id_guard
guard
Stop if id present equals
file_meta
core.storage.data-file-read
Data file read storage
defaults
defaults
Apply defaults
is_image
starts_with
Check if mime type starts with image/
is_pdf
compare
Check if mime type equals application/pdf
readable_kind
any
Check if any of the conditions are true
kind_guard
guard
Stop if readable kind equals
existing_for_file
recued-core.expense-ledger.expense.search
Search expense on expense ledger
existing_rows
default
Apply default
already_captured_count
count
Count items in existing rows
already_captured
compare
Check if already captured count is greater than 0
recapture_guard
guard
Stop if already captured equals
ocr
recued-core.tesseract.image.ocr
Ocr image on tesseract
skip: step.is_image equal false
pdf_text
recued-core.pdftotext.pdf.to_text
To text pdf on pdftotext
skip: step.is_pdf equal false
text_ref
coalesce
Use the first available value from: ocr file ref, pdf text file ref
text_produced
compare
Check if text ref has data
text_guard
guard
Stop if text produced equals
text_persist
core.storage.file.persist
Persist storage file
text_record_id
default
Apply default
text_read
core.storage.data-file-read
Data file read storage
skip: step.text_record_id is_empty
text_decoded
decode_base64
Apply decode_base64
text_body
trim
Trim whitespace from text decoded
text_length
string_length
Apply string_length
has_text
compare
Check if text length is greater than 20
captured_at
date_add
Add 0 seconds to
no_text_note
template
Generate text from a template
no_text_row
recued-core.expense-ledger.expense.create
Create expense on expense ledger
skip: step.has_text equal true
extract
core.ai.extract
AI extracts structured fields from data
skip: step.has_text equal false
defaults_2
defaults
Apply defaults
date_parts
split
Split purchase date into parts
period_raw
template
Generate text from a template
period
ternary
Apply ternary
dup_candidates
recued-core.expense-ledger.expense.search
Search expense on expense ledger
skip: step.has_text equal false
dup_rows
default
Apply default
same_total_rows
filter
Filter by condition
dup_count
count
Count items in same total rows
duplicate_suspect
compare
Check if dup count is greater than 0
confident_enough
compare
Check if confidence is at least setting min confidence
capture_status
ternary
Apply ternary
dup_note
ternary
Apply ternary
expense_row
recued-core.expense-ledger.expense.create
Create expense on expense ledger
skip: step.has_text equal false
expense_id
coalesce
Use the first available value from: expense row record id, no text row record id
card
to_summary
Format results as a summary card
▼
Output
summary
card
Settings
Configurable at install. Defaults shown — change them anytime in Recued.
language
setting
=
eng
record id
setting
=
[object Object]
min confidence
setting
=
0.6
Uses (7 operations)
Files
Data file read storage
core.storage.data-file-read
Pack operation
Search expense on expense ledger
recued-core.expense-ledger.expense.search
Pack operation
Ocr image on tesseract
recued-core.tesseract.image.ocr
Pack operation
To text pdf on pdftotext
recued-core.pdftotext.pdf.to_text
Files
Persist storage file
core.storage.file.persist
Pack operation
Create expense on expense ledger
recued-core.expense-ledger.expense.create
AI models
AI extracts structured fields from data
core.ai.extract
Trust & control
What installing this recipe would let it do. Recued grants these permissions at install — review them there before approving.
What it's allowed to do
Read
Files
core.storage.data-file-read
read-only
Write
Pack
Expense Ledger
external change
Read: expense.search · No approvalWrite: expense.create · No approval
Write
Files
core.storage.file.persist
external change
Read
AI models
core.ai.extract
read-only
2 more aren't listed here. You'll see them at install.