We’re releasing Polyvia 1, as two products:

• Polyvia API: Multimodal Document Retrieval API (for developers of AI agents) - available now.
• Polyvia Platform: Research and Automation Agent over 100K+ multimodal docs (for knowledge workers in enterprises) - coming soon.

We index your unstructured & visual & multimodal docs (PDFs, charts, slides, complex tables, infographics, scans, handwriting, invoices, and more) into multimodal knowledge ontology, with agents running on top for retrieval, research and automation — every answer grounded in a cited source page, in sub-200ms.

Why Polyvia

1. Fast over 100K+ multimodal docs. Agentic, file-by-file search (Claude Code, Claude Cowork, Codex) works only up to ~100 multimodal files — past that it's too slow, and at scale you still need retrieval. Polyvia does sub-200ms search over 100K+ files, every answer grounded in a cited source page.

2. End-to-end — no need for extractors or PDF parsers. When you build large-scale multimodal RAG over a company's files, the only infra available today is visual extractors / PDF parsers (Reducto, LlamaIndex). There's no end-to-end infra for large-scale multimodal document retrieval — until Polyvia: VLM Visual Extractor → Multimodal Knowledge Ontology (mapping all your company's data and processes) → Self-Improving Retrieval Agent.

3. All unstructured, visual and multimodal data inputs in one API. Available now: PDFs, charts, infographics, complex multi-page tables, slides, pictures, handwriting, scans, invoices, audio. Coming soon: video, healthcare scans / EHR, chemical & molecular data, CAD & technical drawings, heatmaps.

What people build with it

• Multimodal RAG inside your own agent — retrieval-as-a-tool over large doc sets.
• Data-room / due-diligence search — query 100+ visual-heavy PDFs jointly (PE, IB, M&A).
• Counterparty & credit monitoring — EBITDA, opex, revenue across hundreds of borrower reports.
• Image-based claim processing — describe claim photos in the context of a policy.
• Cross-engagement slide search — find answers buried in thousands of slides.

Install

pip install polyvia        # Python 3.9+
npm  install polyvia       # Node 18+

Quickstart

Grab a key in Polyvia Platform → Settings → API. Ingest a batch into a group, then ask one question across the whole corpus — answers cite the exact page in each document.

Python SDK

from polyvia import Polyvia

client = Polyvia(api_key="poly_<key>")  # or set POLYVIA_API_KEY

# Ingest a batch into a group, then ask one question across all of it.
items = client.ingest.batch(
    ["q1.pdf", "q2.pdf", "q3.pdf", "q4.pdf"],
    group="FY24 Earnings",
)
for item in items:
    client.ingest.wait(item.task_id)

print(client.query("How did revenue trend across the four quarters?",
                   group="FY24 Earnings").answer)

JavaScript/TypeScript SDK

import { Polyvia } from "polyvia";

const client = new Polyvia({ apiKey: "poly_<key>" });

const items = await client.ingest.batch(
  ["q1.pdf", "q2.pdf", "q3.pdf", "q4.pdf"],
  { group: "FY24 Earnings" },
);
await Promise.all(items.map((i) => client.ingest.wait(i.task_id)));

const answer = await client.query(
  "How did revenue trend across the four quarters?",
  { group: "FY24 Earnings" },
);
console.log(answer.answer);

Scope a query three ways: a single document_id (fastest), a group / group_ids, or the whole workspace (no scope).

More examples

Runnable scripts live in examples/. A few highlights:

Querying across scopes, for example:

# whole workspace · a group (by name) · one document (fastest) · many groups (by id)
client.query("What risks recur across all reports?")
client.query("How did revenue trend?", group="FY24 Earnings")
client.query("Executive summary?", document_id="doc_<id>")
client.query("Compare the deals.", group_ids=["g_<id>", "g_<id>"])

→ Full Developer Docs

Use it from an agent

MCP — connect Claude Code (or any MCP client) to the hosted Polyvia MCP server in one line, so your agent can retrieve over your documents as a tool:

claude mcp add --transport http polyvia https://app.polyvia.ai/mcp \
  --header "Authorization: Bearer poly_<your-key>"

Agent Skills — install Polyvia skills into Claude Code, Cursor, and other agent clients:

npx skills add polyvia-ai/skills

→ MCP docs · Agent Skills

Roadmap

Release log

We update this as we ship — latest first. Full notes at docs.polyvia.ai/versions.

Polyvia-1.1 — Polyvia API · available now

• REST API v1 — ingest, documents, groups, query, usage, rate-limits; async ingestion with task polling and grounded citations.
• Python SDK — pip install polyvia; typed sync and async clients, batch ingestion, idempotent groups, structured errors.
• TypeScript SDK — npm install polyvia; fully typed, ESM/CJS, Node 18+.
• MCP server — claude mcp add --transport http polyvia https://app.polyvia.ai/mcp --header "Authorization: Bearer poly_<your-key>".
• Agent Skills — npx skills add polyvia-ai/skills for Claude Code, Cursor, and other agent clients.
• Visual Document Modalities — Visual Document Intelligence + Audio: charts, graphs & plots, infographics, complex multi-page tables, slides & decks, reports & filings, scanned & photographed pages, invoices & forms, handwriting & annotations, diagrams & flowcharts, photos & images, and audio (calls, meetings, recordings).

Up next

• Polyvia-1.2 — Polyvia Platform — Research & Automation Agent over 100K+ multimodal docs, for knowledge workers in enterprises.
• More modalities (coming soon) — healthcare scans / EHR, chemical & molecular data, CAD & technical drawings, video, heatmaps.
• Polyvia Agents — build your own agent for automating processes on large volumes of multimodal documents.

SDKs & reference

Supported inputs: PDFs · Word/PowerPoint/Excel (DOCX/PPTX/XLSX) · Markdown · text · images · audio. Charts, infographics, complex multi-page tables, slides, scans and handwriting are first-class.

Resources

Runnable snippets (Python, TypeScript, raw HTTP, MCP, agent-tool) live in examples/ — see the examples guide. See also CHANGELOG · CONTRIBUTING · SECURITY.

New to Polyvia? See what it does at polyvia.ai, or start free at app.polyvia.ai.

📚 Docs · 🖥️ Platform · ✉️ mateusz@polyvia.ai · senyao@polyvia.ai

License

© 2026 Polyvia. All rights reserved.