Skip to main content
Polyvia ingests text, visual, and audio source material into the same unified knowledge graph. You don’t need to pre-classify or pre-convert your files — drop in whatever your team produces and Polyvia routes each file through the right parser automatically.

​
Modalities

Live today — Visual Document Intelligence + Audio, ingesting now:
  • Charts
  • Graphs & plots
  • Infographics
  • Complex, multi-page tables
  • Slides & decks
  • Reports & filings
  • Scanned & photographed pages
  • Invoices & forms
  • Handwriting & annotations
  • Diagrams & flowcharts
  • Photos & images
  • Audio (calls, meetings, recordings)
Coming next:
  • Healthcare scans / EHR
  • Chemical & molecular data
  • CAD & technical drawings
  • Video
  • Heatmaps

​
Overview

Documents

PDF, DOCX, PPTX

Text

TXT, Markdown

Images

PNG, JPG, screenshots, scans

Audio

WAV, MP3, M4A, and more

​
Documents

Reports, filings, contracts, research papers, scanned books, invoices. Polyvia reads everything on the page — not just the text layer:
  • Text — body text, headings, footnotes, and multi-column layouts
  • Tables — including complex, multi-page, and nested tables
  • Charts & graphs — bar, line, pie, scatter, and combo charts read back as data
  • Infographics & diagrams — flowcharts, org charts, schematics, maps
  • Figures & images — embedded photos, logos, and screenshots
  • Scanned & photographed pages — OCR over image-only PDFs
  • Handwriting — handwritten notes and annotations
  • Forms & invoices — fields, line items, stamps, and signatures
Page numbers and layout are preserved for citations.Extensions: .pdf
Memos, proposals, internal docs. Headings, lists, and tables are preserved so structure-aware queries (e.g. “what does section 3.2 say about…”) work correctly.Extensions: .docx, .doc
Pitch decks, board decks, training material. Slide order is preserved and per-slide visuals (charts, images, diagrams) are extracted alongside speaker notes.Extensions: .pptx, .ppt

​
Text

Logs, transcripts, notes, raw exports. Indexed line-by-line with no formatting loss.Extensions: .txt
READMEs, wikis, internal docs, AI-generated reports. Headings, code blocks, and lists are parsed natively, so structural citations point to the right section.Extensions: .md

​
Images

Product photography, architecture diagrams, whiteboards, screenshots, scanned receipts. Polyvia runs visual understanding to extract text, structure, and entities — citations point to the image with bounding boxes when applicable.Extensions: .png, .jpg, .jpeg, .gif, .webp, .bmp

​
Audio

Sales calls, interviews, podcasts, meeting recordings. Polyvia transcribes with timestamps and speaker turns — citations link directly to the utterance, and queries can return clickable timestamps that seek the player to the cited moment.Extensions: .wav, .mp3, .m4a

​
How ingestion works

Regardless of format, every file flows through the same pipeline:
  1. Parse — format-specific parser extracts raw content (text, page layout, frames, transcript).
  2. Extract — VLM and LLM passes pull out facts, tables, charts, and entities.
  3. Index — facts are linked into the ontology graph with source provenance (document, page, bounding box, or timestamp).
  4. Query — agents can search across all formats in the same call; citations point back to the exact source location.

​
Uploading

You can ingest any supported format through any of our interfaces:

Polyvia Platform

Drag-and-drop upload — no code required.

REST API

Single or batch upload via /api/v1/ingest.

SDKs

Python and TypeScript SDKs with batch helpers.
Need a format we don’t list? Video, HTML, and email ingestion are on the roadmap. Email senyao@polyvia.ai and tell us what you’d like to throw at Polyvia.