How to use
- Paste text or open a TXT, DOCX or PDF file.
- Choose Key sentences (instant) or AI summary, plus length and format.
- For the AI summary press Download model once, then Summarise; long texts are summarised section by section and the section summaries are combined.
Worked example
On the 8-sentence sample news story, Key sentences at Medium length picks the 5 most central sentences as bullets, quoted word for word. A 2,400-word text of 20 paragraphs is split into 5 sections of up to 600 words for the AI summary, and the five section summaries are summarised again into one.
Supported formats and limits
| Input | Text, PDF (with a text layer), DOCX, TXT |
|---|---|
| Output | Summary (bullets or paragraph) |
| Limits | Key sentences need no download. The AI summary downloads DistilBART-CNN-6-6 (about 286 MB) from Hugging Face once, cached afterwards; long texts are summarised in sections of about 600 words and then combined. Up to about 60,000 words. Your text never leaves this tab. |
| Engine | TextRank extractive summary (on device, instant) or DistilBART-CNN-6-6 (Apache-2.0) via transformers.js in a Web Worker; map-reduce over 600-word sections |
Limitations
- AI summaries can omit or misstate details, and each combining stage loses more; use them as a guide, not a replacement.
- The model was trained on English news and works best on article-like prose.
- Key sentences only quote the text; they cannot rephrase or merge points.
Questions
What is the difference between the two modes?
Key sentences uses TextRank to pick the most central sentences and quotes them word for word, with no download. AI summary uses DistilBART-CNN-6-6 (Apache-2.0, about 286 MB, downloaded once from Hugging Face) to write new sentences.
Can I trust the AI summary?
Use it as a guide. It is machine-generated by a small model trained on English news, and it can leave out or misstate details. On long texts each combining stage loses more, so check anything important against the original.
How long a document can it handle?
Up to about 60,000 words. For the AI summary, text is split into sections of about 600 words, each is summarised, and the section summaries are summarised again. Scanned PDFs without a text layer need OCR first.
Privacy
On-device model. Processing runs in this browser. The open-source model files are downloaded once from the model host (Hugging Face) and cached; your content is not uploaded.
- Hugging Face: Only after you press Download model: requests for the model files (DistilBART-CNN-6-6, about 286 MB, only for the AI summary) from huggingface.co and its CDN, which see your IP address. Your text and files are never sent.
See the privacy policy for how toolsdocks handles data.