Reducto Agentic Document Platform

Reducto, Inc. is a Delaware corporation based at 77 Geary Street, San Francisco, backed by a16z, Benchmark, First Round Capital and Y Combinator. It provides document ingestion infrastructure for AI teams and enterprises that need reliable data from complex PDFs, scans and spreadsheets before feeding them to language models or business systems.

The platform combines custom vision models with frontier LLMs. Parse converts more than 30 file types into structured, citation-ready JSON with tables, charts and handwriting; Extract returns typed JSON with bounding boxes; Split and Classify segment and route documents; Edit writes data back into PDF and DOCX; and Studio builds pipelines without code. Standard OCR detects more than 60 languages automatically, including Vietnamese, Thai, Indonesian, Malay and Khmer, and an agentic mode uses a vision model to correct difficult text.

• Parse: converts PDFs, scans, spreadsheets and slides into structured JSON with citations.
• Extract: returns schema-based typed JSON with bounding boxes.
• Split and Classify: segments long files and routes documents with plain-language taxonomies.
• Edit: fills extracted data back into finished PDF and DOCX files.
• Multilingual and agentic OCR: auto-detects 60+ languages and corrects handwriting and poor scans.
• Developer tooling: Python and Node SDKs, CLI, MCP server and no-code Studio.

Vietnamese banks can parse long credit files and financial statements into JSON for RAG-based credit assistants. Insurers can extract fields from mixed Vietnamese and English claim documents. Legal and audit firms can split and classify large contract bundles before review.

Outputs carry citations and bounding boxes, so every extracted value can be traced back to the page. Mixed-language documents need no language configuration. An independent LongExtractBench test in June 2026 ranked Reducto Deep Extract first with 99.6% precision and recall on 225 long documents. Deployment ranges from managed cloud to VPC, on-premises and air-gapped setups.

Retrieval-augmented generation and AI agents depend on clean document data, which has made document parsing a core layer of the AI stack and attracted large venture rounds in 2025. Southeast Asian banks, insurers and AI startups building assistants over Vietnamese, Thai or Indonesian documents need parsers that handle local scripts and mixed languages.

Reducto, Inc. was co-founded by Adit and Raunak and is headquartered in San Francisco. Its October 2025 Series B of 75 million USD was led by a16z, with Benchmark, First Round Capital, Y Combinator and BoxGroup, for 108 million USD in total funding. It reports more than one billion pages processed, SOC 2 Type II audit, HIPAA-compliant processing with BAAs and zero data retention by default.