Cleaners
What are Cleaners?
Cleaners are AI-powered enrichment configurations that re-process and normalise specific fields in your extracted data. Where a flow extracts raw values from a document, a cleaner applies rules, validation, and AI reasoning to clean or enrich those values.
A flow extracts data from a document. A cleaner takes that data and cleans or enriches it — for example normalising date formats, correcting misspellings, or classifying values into categories.
Cleaner Fields
Each cleaner defines one or more fields to process. For every field you can configure:
- Extraction hints — examples and patterns that guide the AI
- Allowed values — restrict output to a fixed list of options
- Sub-fields — break a field into nested child fields
- Regex patterns — validate or transform the extracted value
The more examples you provide in the extraction hints, the more accurately the AI will clean your data. Include edge cases and common variations.
Creating a Cleaner
Use the multi-step creation wizard to set up a cleaner:
- 1Go to Cleaners in the main navigation and click "New Cleaner"
- 2Give your cleaner a descriptive name
- 3Upload sample documents — the AI will discover candidate fields
- 4Review and customise the discovered fields, adding hints and constraints
- 5Link the cleaner to one or more flows (optional)
- 6Save and activate
Running a Sweep
A “sweep” is one execution of a cleaner against a batch of documents or extracted data. You can trigger sweeps manually or link them to flow runs so they fire automatically.
Open the cleaner's detail page, click “Run Sweep”, and select the data to process.
When a cleaner is linked to a flow, it can be set to run automatically after each extraction run completes.
Viewing Sweep Results
After a sweep completes, open it from the Sweeps history on the cleaner's page to see:
- Per-field cleaned values and confidence scores
- Number of records processed and any failures
- Credit usage for the sweep
- A comparison of input vs output values