nuvo

The mapper compares imported headers and values against your target model, then proposes a match for each column. Names, labels, and any synonyms you add in alternativeMatches all count toward the score. Mapping runs in DEFAULT mode, where AI fills gaps on every execution, or EXACT mode, where only confirmed columns pass through. The vendor lists support for more than 50 languages, which matters when customer files arrive in German, French, or Japanese.

A target data model defines the columns your destination expects: labels, technical keys, types, and validation rules. Types go beyond text and numbers to dates, timestamps, booleans, categories, country codes, and currency codes. Each column can require a mapping or reject values that break a format, and category columns support option mapping with a similarity threshold. Get this layer right and the rest of the pipeline has a fixed shape to aim at.

Input and output connectors cover HTTP(S), (S)FTP, AWS S3, Azure Blob Storage, Google Cloud Storage, email, and manual upload. Files can be CSV, Excel, TSV, XML, or JSON. A CSV import reads the first row as headers, and FTP connectors can rename or delete a file after pickup, and output file names can include timestamps or the original name. HTTP uploads stop at 4.7 MB, and a workbook with several sheets only sends the first one.

Transformations use spreadsheet syntax, so IF, REPLACE, CONCATENATE, TEXT, and date functions look familiar to anyone who has built a sheet formula. A prompt field drafts the formula from a plain description, which shortens the trial and error. Option mapping handles category values separately, matching "DE" and "Germany" to one stored option. Each transformation is stored per column, so future runs repeat it without manual work.

Schedules run hourly, daily, weekly, or monthly with an interval, plus a start date and an optional end date. A weekly run at 10 a.m. CET is a normal setup. Event-based triggers work the other way: a file sent to an HTTP endpoint or an email address starts the pipeline as it lands. Both modes can feed the same target model, which helps when some customers batch files and others send them one at a time.

Every run appears in one list with status, timing, and the number of rejected cells. The error threshold is a percentage, so a run with 8% bad cells can pass when the limit sits at 10%. Failed executions can be fixed once and rerun, or the pipeline config can be updated so the change applies to future runs. Permissions can be split by component, so support staff fix executions while engineers own connectors.

The embeddable importer ships as React, Angular, Vue, and VanillaJS components covering pipeline creation, connector setup, target model editing, file upload, and execution repair. A REST API covers pipelines, executions, connectors, target models, users, and sub-organizations, so you can build your own interface and keep the engine headless. Sub-organizations give each customer a separate space, with public or private components deciding what gets shared.