Polyvia - Product Hunt

Multimodal Document
Retrieval API

Join Waitlist
polyviaStarDocumentation

Products

API for developers, Platform for teams

Polyvia API

Try API

Multimodal Document Retrieval API
— for Developers of AI Agents

Python$pip install polyvia
Typescript$npm install polyvia
Agent Skills$npx skills add polyvia-ai/skills
MCP$claude mcp add --transport http polyvia https://app.polyvia.ai/mcp --header "Authorization: Bearer poly_<your-key>"
# pip install polyvia
from polyvia import Polyvia
client = Polyvia(api_key=API_KEY)
# Ingest & query scoped to a group
client.ingest.batch(["q4.pdf", "10k.pdf"], group="Q4 Earnings")
answer = client.query(
  "Compare EBITDA across all filings",
  group="Q4 Earnings",
).answer
Python SDK →TypeScript SDK →MCP →Agent Skills →

Polyvia Platform

Try Platform

Search and Exploration over multimodal docs
— for Knowledge Workers in Enterprises

Interactive Exploration of Multimodal Knowledge Ontology
Platform Docs →

Use Cases

How teams put Polyvia to work

EXTRACTED FACTS147 facts

Data-Room Due Diligence

Surface every revenue, churn, and customer-concentration fact across a target's decks and statements.

AAPLMSFTGOOGMETAAMZNNVDA+42%Gross margin · FY25

Cross-Filing KPI Comparison

Compare a single metric across 500+ counterparty filings in seconds.

Acme CorpDSCR 1.42 · OKGlobex IncDSCR 1.08 · WatchInitech LtdDSCR 0.91 · BreachSoylent CoDSCR 1.71 · OK

Counterparty Credit Monitoring

Flag covenant breaches and exposure shifts across 100+ borrower reports automatically.

damage: hailseverity: highauto-routed

Image-Based Claim Processing

Extract damage type, severity, and location from claim photos; auto-route to adjusters.

Why Polyvia

Advances over current solutions

Fast at scale

Fast over 100+ multimodal docs

Agentic, file-by-file search (Claude Code, Cowork) is too slow past ~100 multimodal docs — at scale you still need retrieval. Polyvia answers in sub-200ms over 100K+ files.

File-by-fileminutes
Polyvia<200ms

End-to-end

No need for extractors or PDF parsers

One end-to-end solution for large-scale multimodal document retrieval — not a stack of vendors (Reducto, LlamaIndex) to stitch together.

VLM Visual Extractor→Knowledge Ontology→Self-Improving Retrieval Agent

Enterprise & secure

Data never leaves your systems

On-prem deployment of Polyvia agents inside your own cloud / VPC, via the Polyvia for Enterprise offering — your documents stay within your perimeter.

Encrypted
Your VPC · on-premPolyvia

What Polyvia Can Do

Production-ready from day one

Audit-ready answers

Every answer traced back to source

Which segments show the fastest growth?
Cloud services led growth (+42%).cite: 10-K p.42
10-K · p.42 ¶399.8%

Cloud services led growth at +42%, driven by enterprise contract expansion.

99.8% citation coverage

Built for scale

From 5 files to 100K+ documents

Throughput · last 24h42K / hr
0
Documents indexed
sub-200ms
Query latency
0
Facts per corpus
0
Extraction confidence

Integrations

Works with your stack

AWSAWS S3
SnowflakeSnowflake
GoogleGoogle Drive
SharePoint
CRM
ERP
NotionNotion
Dropbox
Slack
CursorCursor
ClaudeClaude
OpenAICodex

Multimodal Ingest

Every unstructured, visual and multimodal input

Visual Document Intelligence
+42%

Charts

MetricQ4Growth

Complex tables

Infographics

Slides & reports

INVOICE$12,847Sig.

Scans, Invoices & Handwriting

Pictures

Standard

Text

Multimodal Document Intelligence
02:14 → 02:31

Audio & Transcripts

Video

Coming soon
C₈H₈O₂

Molecular & Chemical

Coming soon
120mm⌀22

CAD & Drawings

Coming soon

Geospatial & Heatmaps

Coming soon

Read · See · Listen

Pinpoint precision in every format

ReadDocuments

Cited to the paragraph

Q4 revenue rose to $24.8B, up 14% year-over-year.

Cloud services led growth at +42% Consumer hardware was flat.

Net retention reached 118%, the highest in eight quarters…

cite · 10-K · p.42
SupportsPDFDOCXMDTXT
SeeSlides

Read the chart, not the caption

+42%cite · deck p.7 · chart 2
SupportsPPTXPDF
SeeImages

Bounding-box precision

cite · fig 4 · (140,16)→(230,104)
SupportsPNGJPGWEBP
ListenAudio

Cited to the second, not the recording

00:0002:1402:3105:00
cite · 02:14 → 02:31 · “…retention reached 118%”
SupportsWAVMP3M4A

End-to-End Pipeline for Multimodal Document Intelligence

Polyvia Engine

01

VLM Visual Extractor

Quarterly RevenueEXTRACTED · JSONperiod: Q4 2025revenue: $24.8Bgrowth: +42%source: deck p.7chart: barconfidence 99.8%

SOTA visual document extractor and parser. Fine-tuned VLM-OCR pipeline for the hardest visual and multimodal inputs. Extracts actual data points — not 300-token descriptions.

VLMOCRfine-tuned
02

Multimodal Knowledge Ontology

Knowledge graph for large-scale visual file search. Disambiguates extracted facts into unique entities; connects them across the corpus. Single source of truth, cross-document reasoning across 100K+ files.

graphentity-linking100K+ files
03

Self-Improving Retrieval Agent

Which segments show fastest growth across all filings?DECOMPOSEDrevenue · Q4segments · YoYgrowth · citeLLM-as-Judge · rerankCloud +42% (10-K p.42)cite · cite · cite ✓218ms

Query decomposition + iterative retrieval + LLM-As-A-Judge. Sub-200ms graph search across 100K+ documents. Every answer grounded in visual citations. Learns which retrievals lead to successful generations.

sub-200mscitedself-improving

Start building with Polyvia

Get access to Polyvia, and stay up to date with updates!
Enter your email to join.

We're rolling out access weekly.