オンデバイス AI · WebGPU

どんな PDF も 読まずに理解。

要約・要点・固有名詞抽出をすべてブラウザ内で。サーバーを経由しないオンデバイス LLM。

GPU を検出中…
SmolLM2 · Qwen2.5
100% プライベート
PDF 要約の特長

賢く読む。アップロードなし。

ローカル LLM が文書を読み、構造化されたインサイトを数秒で提供。

100% プライベート

文書はデバイスから出ません。推論は WebGPU でローカル実行。

2 つのオープンモデル

SmolLM2 1.7B か Qwen2.5 1.5B を選択。小型・高速・高性能。

長文対応

長い PDF は部分ごとに要約し、最後に統合して仕上げます。

構造化出力

Markdown 形式の要約・要点・エンティティ一覧をそのままコピー。

How it works

Three steps. Zero servers.

  1. 01

    Upload a PDF

    Text is extracted page by page in your browser via pdf.js. Nothing is uploaded.

  2. 02

    Pick a task & model

    Summary, key points, entity extraction, or your own question — answered by SmolLM2 or Qwen2.5 running locally.

  3. 03

    Read the brief

    Answers stream in as they are generated, ready to copy as Markdown.

Complete guide

About the PDF summarizer

A free tool that summarizes PDFs and answers questions about them with a language model that runs on your device — no API keys, no uploads, no per-page fees. It is aimed at students, researchers, and professionals who need the gist of long papers, reports, and documentation quickly.

How it works

pdf.js extracts the text layer locally, then the document is chunked and mapped-reduce style summarized by SmolLM2, a small language model running in the browser via Transformers.js. In QA mode, the most relevant passages are retrieved and answered from the document itself. Progress streams live, and results are copy-ready or exportable as Markdown.

Limits & requirements

On-device models are small by design: they summarize well and answer factual questions grounded in the text, but they are not a frontier chatbot for open-ended reasoning. Only text-based PDFs are supported (scanned PDFs need OCR). Chrome or Edge with WebGPU is strongly recommended; model weights download once and are cached.

Privacy

Papers, contracts, and notes never leave your machine. Summarization runs locally after the one-time model download, so documents that would be reckless to paste into a cloud LLM can be summarized safely.

サポート

よくある質問

いいえ。テキスト抽出(pdf.js)も生成(言語モデル)も、すべてブラウザ内で実行されます。モデルは Hugging Face から一度ダウンロードされ、ローカルにキャッシュされます。