CSV import

Import (in the sidebar, from the bundled plugin wickermoney.import-csv) brings a bank or card statement in as transactions. You choose the account and the file, tell it which column is which, look at what it made of your choices, and only then does anything get written. The mapping is remembered, so the second import from the same bank needs no setup, and any import can be undone.

Added in 0.1.0.

The four steps

1. File and account

Pick the account under Import into and choose a .csv file. An import always lands in one account; archived accounts aren't offered. A file with no header row is refused: export it again with column names.

The file is read in your browser, so you see a preview before anything is sent to the server for checking. Its text is read as UTF-8.

What the reader copes with:

• Delimiters: comma, semicolon, tab or pipe, chosen from the header row.
• Quoted cells with commas, line breaks or doubled quotes inside, CRLF or LF line endings, and the byte-order mark Excel adds.
• Ragged rows: short rows are padded and long rows trimmed to the header's width, rather than refusing the file.

2. Columns and formats

Setting	What it does
Date and Description	Required. Which column holds the date and the merchant text
Date format	How every date in the file is written. Choose from MM/DD/YYYY, DD/MM/YYYY, YYYY-MM-DD, MM-DD-YYYY, DD-MM-YYYY, DD.MM.YYYY and YYYY/MM/DD
Amount columns	One signed amount column, or separate debit and credit columns
Amount, or Debit (money out) and Credit (money in)	The column or columns holding money
Notes	Optional. A memo or reference column
Transaction ID	Optional. The bank's own id for each transaction (see Duplicates)
Remember as	A name for this layout, such as "Chase Checking". Starts as the file's name
Flip every sign	For files that show spending as a positive number

The page suggests columns from the header names, and always shows them in dropdowns for you to confirm.

The date format is never guessed from the data. 03/04/2026 is a valid date under two formats and means different months in each. The picker starts at MM/DD/YYYY if your browser's language is US English and DD/MM/YYYY otherwise, and the preview table under it shows the first eight readable rows as they will be saved. Change the format and the table changes at once, so a wrong choice is plain before you commit. A two-digit year is read as 20xx, a time after the date is ignored, and a date that doesn't exist (31/02) is an error, not a rollover to March.

Amounts can have thousands separators, a currency symbol, parentheses or a trailing minus for negatives, and a European decimal comma (a lone comma counts as decimal only when one or two digits follow it, so 1,234 is one thousand two hundred thirty-four). More than four decimal places is an error. With separate columns, exactly one of Debit and Credit must be filled on a row. Money out is negative, matching the rest of the app.

A row that can't be read, whether for a missing or impossible date, an empty description or an amount that isn't money, is skipped, not guessed at. A banner above the preview says how many and gives the first reason.

3. Review

Press Check for duplicates. Nothing is written yet. The page shows counts for the whole file:

Count	Meaning
New	Will be imported
Already imported	An exact match on the bank's transaction id, or an id repeated in the file. Skipped, with no decision for you
Needs review	Looks like a transaction you already have. Listed below, not imported unless you tick it
Unreadable	Rows that couldn't be mapped (shown only when there are some)

Under Possible duplicates, each flagged row is listed beside the transaction it resembles. Tick Import on the ones that really are new. The list is shown a page at a time, and Show more asks for the next page. Rows you haven't ticked aren't imported, whether or not they are on screen.

The button says exactly what will happen: Import N transactions, where N is the new rows plus those you ticked.

4. Imported

A line tells you what was written: how many were imported, how many were already present, how many possible duplicates the file contained and how many rows were unreadable. Import another file starts over.

Duplicates

There are two mechanisms, and the difference between them is the point.

• With a transaction ID column, a match is a fact. A row whose id is already in that account, or that repeats an earlier id in the same file, is skipped without asking. Two coffees bought the same day with different ids aren't flagged against each other. The database enforces this too: a row that lost a race to another import with the same id is counted as skipped, not as a failure.
• Without one, all that's left is a guess. A row is flagged for review when an existing transaction in the same account has the same amount, a date within one day, and a similar description. Descriptions are compared ignoring case and punctuation, and count as similar if they're equal or one is the start of the other (the shorter at least six characters), so AMAZON MKTPL matches AMAZON MKTPL 4XJ22. Only transactions already in the account are compared: two identical rows inside one file, with no ids, are not flagged against each other.

A guess has to be a question, because two identical coffees on one day look the same as one coffee imported twice. Dropping the second silently would lose a real transaction.

If your bank gives you an id column, map it: it makes re-importing an overlapping statement safe and quiet. Both checks look at the account's transactions from two days before the file's earliest date to two days after its latest. An id that matches something older is still caught, by the database, when you commit.

Remembered mappings

When you import, the mapping is saved under the name in Remember as. A name you reuse replaces the old mapping, ignoring case. Pick a file later and the page looks for a saved mapping whose columns all appear in the file's headers (ignoring case), applies the first that does, and says "Using your saved mapping for Name". It matches on the columns, not the file name, since banks name exports activity (3).csv. Everything it filled in is still in the dropdowns to change.

Categories and rules

Imported transactions are filed by your rules as they are written, using the same rule order as everywhere else. A transaction no rule matches comes in uncategorized, and you work through those on the Transactions page with Only uncategorized. Rows filed this way carry the rule tag on the Transactions page.

An import never creates a transfer. A card payment shows on both statements as two ordinary transactions. See A transfer that isn't recorded as a pair.

Undoing an import

Recent imports at the bottom of the page lists the 25 latest, each with the account, the source, the file name and the counts. Undo removes what that import created. It uses a record of exactly which transactions the batch wrote, so a transaction you entered by hand that looks the same is safe.

Undo keeps a transaction you've changed since the import, and tells you how many. Changed means you edited it, split it, set its category by hand, or turned it into a transfer. A category a rule set on import doesn't count as a change. An import can be undone once.

Limits

• 8 MB and 100,000 rows. The file can't be larger than 8 MB. A file with more than IMPORT_MAX_ROWS data rows, counting rows that can't be read, is refused with a message that names the limit, before anything is written. Split the file and import it in parts. The default of 100,000 is measured, with the reasoning, on the configuration page. An import runs under DB_LONG_STATEMENT_TIMEOUT, not the usual 30 seconds.
• One import into an account at a time. Imports into the same account take turns, so the second sees what the first wrote. Imports into other accounts, by you or by other people on the instance, run alongside, each holding a database connection for its commit.
• One commit is one batch. Everything is written in a single transaction, or nothing is. A repeat of the same request, such as a double click or a retry after a network error, writes nothing the second time.
• CSV only. There is no OFX, QIF or PDF reader, and no connection to a bank.
• UTF-8. The file is read as UTF-8, so a file in another encoding may show garbled descriptions.

For the endpoints, see API reference. Turning the plugin off keeps your saved mappings and import history (Plugins).