Reducto

Reducto was co-founded by Adit Abraham and Raunak Chowdhuri in San Francisco and joined Y Combinator’s Winter 2024 batch. Both founders studied computer science at MIT.⁠[1]

The company announced $8.4 million in funding led by First Round CapitalFirst Round CapitalVenture capital firm.Open the full entry, with participation from YC and BoxGroup. It traced its origins to enterprise retrieval-augmented generation projects, where ingesting complex PDFs and spreadsheets had become a bottleneck.⁠[2]

Reducto announced a $24.5 million Series A led by Benchmark, bringing total funding to about $33 million. It also introduced Agentic OCR, using multiple vision-language-model passes to review parsing output and correct errors in complex layouts, equations and positional information.⁠[3]

The company announced a $75 million Series B led by a16zAndreessen HorowitzAndreessen Horowitz (a16z) is a venture capital firm founded in 2009 by Marc Andreessen and Ben Horowitz. It invests from the seed stage through the growth stage and provides operational support to its portfolio companies.Open the full entry, with Benchmark, First Round Capital, YC and BoxGroup participating. Total funding reached $108 million. The financing supported model research, product expansion and enterprise adoption; Reducto reported over one billion pages processed at that time.⁠[4]

Reducto launched its Office plugin for ChatGPT and Codex, supporting Word, PowerPoint and Excel editing while preserving formatting, comments and tracked changes. Users can inspect results and continue editing in a live editor. At launch it was free with usage limits and required a connected Reducto account.⁠[5]

The platform serves AI companies and document-heavy enterprises by combining parsing, field extraction, splitting, classification and editing. Reducto names Harvey, Scale AI and Vanta among its users, with supported material spanning PDFs, scans, spreadsheets, slides and handwriting.⁠[6]

Parse converts documents into structured JSON through OCR, layout detection, table reconstruction, figure summaries and semantic chunking. Blocks carry types, page locations and confidence scores, allowing retrieval and knowledge-base applications to trace results to source paragraphs, table rows or figures.⁠[7]

Extract returns fields using a user-defined schema for material such as invoice items, contract dates and financial statements. Citations can include page numbers, bounding boxes and source text. Deep Extract adds repeated checks and re-extraction against the original for long tables, missing rows and inconsistencies.⁠[8]

Split identifies section boundaries from natural-language descriptions and returns page ranges. Partition keys, such as account numbers, distinguish repeated sections. Annual reports, bundled statements and intake packets can be separated before downstream extraction, with low-confidence segments routed for human review.⁠[9]

Studio provides source-and-output previews for tuning OCR, chunking and extraction fields. Applications call the resulting workflow through a stable Pipeline ID. Versioning, rollback and configuration comparisons let operational teams and developers inspect citations, tables and extracted results together.⁠[10]

Reducto Editor provides an embeddable UI and SDK for teams adding DOCX, PPTX and XLSX editing to their own products. Edits write into the file’s XML, with standard Word revisions and comments. The engine can run in browsers or customer servers and exports regular Office files.⁠[11]

The published policy sets Growth-and-above API data to expire within 24 hours, excluding Studio jobs and explicitly persisted results, and says those tiers’ data is not used for training. Its “zero retention” term therefore does not mean immediate deletion for every workflow. Reducto also reports completing SOC 2 Type I and Type II processes.⁠[12]