OpenRefine alternative with nothing to install

Open-ended exploration. Faceting a column to see every distinct value, then reshaping the data with a transformation you write yourself, is something a proposal-based tool cannot match. If your cleaning job is unusual and you enjoy the tooling, OpenRefine is a good place to spend an afternoon.

It is also better at one specific job: making every spelling of a value in a column the same. Its key collision methods (fingerprint, n-gram fingerprint, Metaphone3, Cologne, Daitch-Mokotoff and Beider-Morse) and its nearest neighbor methods (Levenshtein and PPM) group spellings like Acme Corp, ACME CORP. and Acme, Corp so you can pick one. DataUntangler does not do that across separate records, and it is free and open source, which matters when a budget line is the obstacle.

We built an eight row contact file with the faults that send people looking for an OpenRefine alternative in the first place: four contacts, each in the file twice. Ana Brooks appears once at Acme Corp and once at ACME CORP. with the same email. Ben Ito is at Acme Corporation and at Acme, Corp with two different addresses at acmecorp.com. Carla Mendes works for an Acme Plumbing in Tulsa and was also entered as Acme Corp. Dev Shah is at Pinecrest Partners and Pinecrest Partners LLC, with two emails.

In OpenRefine you would open the Company column and run fingerprint clustering. Fingerprinting lowercases the value, strips punctuation and sorts the words, so Acme Corp, ACME CORP. and Acme, Corp all collapse to the same key and are offered as one cluster. That includes the Acme Corp typed against Carla, a different business in a different state, because clustering looks at the cell and not at the rest of the row. Acme Corporation only joins under a nearest neighbor method. When you accept, the cells change. The eight rows are still eight rows, and removing the duplicates is a second job you do yourself.

Both tools leave you a decision, and they are different decisions. In OpenRefine you decide which spelling wins for a whole column, and you can easily give Carla a company she never worked for. In this engine you decide whether two rows are the same contact, and the cases that are not certain carry a flag. Carla is the honest example on our side: her two rows share an email, so they merge, and the survivor keeps Acme Corp. The audit log records that Acme Plumbing was the value on the removed row, so it is recoverable, but it is not kept in a column of its own the way a second email is.

What the engine did not do is just as important. After the merge, Ana stays at Acme Corp and Ben stays at Acme Corporation. They are different people at what is probably the same firm, and this engine does not rewrite the company on one record to match the company on another. If one spelling of every company is the goal, run OpenRefine clustering on the column after the duplicates are gone, or do it in the CRM.

Time to a result, and the ability to hand the result to someone else. There is no runtime to install on a locked-down laptop, no expression language between you and the fix, and the output includes an audit log written for a person who was not there when you cleaned the file. The wider comparison sits on data cleaning software, and if the file is a CRM export on its way back in, CRM data cleansing covers the import side.

OpenRefine keeps an undo and redo history for each project, and you can extract it as JSON to replay the same operations on another file. That is a good tool for the person who did the cleaning. It is not something a finance controller or a CRM admin reads. A CSV with the row, the column, the original value, the new value, the rule and the decision is.

The people who leave are usually not data specialists. They are an operations manager who inherited a spreadsheet of customers, a marketer merging two event lists, or an analyst whose IT team will not approve a Java application on a company laptop. They tried OpenRefine because it came up first, spent an hour on facets and clusters, and still had the duplicate rows at the end. For them, a list of proposed merges with a reason next to each one is the whole product, and the data cleaning tool on the homepage is the same engine as the demo on this page.

People who should stay are researchers, journalists and data librarians who reshape unusual data, reconcile values against Wikidata, or write GREL every week. OpenRefine is built for them and it is very good at it. If that is you, keep it and use this page as a second opinion on duplicates only.