Reducto

Reducto turns PDFs, scanned pages, spreadsheets and slides into structured data for AI applications. Its customers include AI startups and enterprises working with financial, legal and healthcare documents; the company names Harvey, Scale AI and Datasite among its users. Its Parse product preserves document structure, while Extract retrieves fields specified by an application’s schema, such as information from invoices or financial disclosures. These tools give developers a way to bring document contents into downstream workflows.

Reducto’s document-processing focus grew out of an earlier effort to give language models long-term memory. Users asked whether the system could also process uploaded files, and that ingestion work revealed a more immediate customer need. Co-founder Adit Abraham described the transition in a founder interview: failures in a cloud document-processing service pushed the team to develop its own layout understanding. Reducto initially served engineering teams through an API, then expanded toward business teams handling document-heavy back-office work.

Its technical approach combines computer vision with vision-language models. The r-1 parsing model interprets layout, text, tables, figures and formatting together, returning structured output with bounding-box citations that locate information on the original page. This preserves context that matters when a table’s arrangement or a document’s formatting carries meaning. Enterprises can also run Reducto within their own infrastructure to meet data-residency and security requirements.