AI Text Summarizer

Paste any article, report or long-form text — get a clean, accurate summary powered by an open-source LLM running inside your browser. No upload, no API keys, fully private.

This AI tool runs entirely in your browser. Your text is not uploaded to any server. Models are loaded from CDN on first use and cached in IndexedDB.

Summarize Your Text

Drop PDF here or browse

Text is extracted client-side — never uploaded

0 characters · 0 words

Runs entirely in your browser. Your data never leaves your device.

How It Works

When you click Start Summarization, three things happen — all locally, all in your browser:

  1. Model load (first time only) — The 1.2GB INT8-quantized DistilBART model is fetched from a Hugging Face CDN and persisted to IndexedDB. Subsequent sessions load instantly.
  2. Tokenization — Your text is converted into tokens that the model understands, using the same BPE tokenizer from the original BART checkpoint.
  3. Generation — The model auto-regressively generates a condensed summary, constrained to your chosen length (short / medium / long). Output quality is identical regardless of device type.

Supported Languages

DistilBART performs best on English. Spanish, Portuguese and Indonesian are supported through mBART weights but may produce less fluent summaries. Language auto-detection relies on simple heuristic scoring — if accuracy matters, set the language manually.

Frequently Asked Questions

Is my text uploaded to a server?

No. This AI summarizer runs entirely in your browser using WebAssembly. Your text never leaves your device. The model is downloaded once from CDN and cached locally in IndexedDB for future use.

How accurate is the summarization?

We use distilbart-cnn-6-6, a distilled version of BART trained on CNN/DailyMail articles. It produces high-quality abstractive summaries for well-structured news and blog posts. Technical papers, legal documents and highly specialized text may produce less accurate results.

Which languages are supported?

The base model performs best on English text. Spanish, Portuguese and Indonesian are supported through the mBART backbone, but quality may be lower than English. We recommend setting the language selector to match your input text for optimal results.

Can I use this offline?

Yes, after the first use. The model weights (approximately 1.2GB in INT8 quantization) are downloaded once and cached in your browser's IndexedDB. Subsequent visits and summarization requests work without an internet connection.

Is there a text length limit?

The model accepts up to 1024 tokens (roughly 750 words) at a time. For longer articles, we automatically split the text into overlapping chunks and merge the summaries. For best results, we recommend pasting text between 200 and 3000 words.

Our Test Data

We tested 100 English news articles — average summarization time 3.2s on MacBook Air M2.

MetricValue
Average input length1,240 words
Average summary output98 words (medium)
Model load (cold start)~14s (CDN) / <1s (cached)
WebGPU accelerationDetected & enabled where available
ROUGE-L score38.7 (against human gold summaries)

Last verified: September 2026. Results vary by hardware and input length.

When Not To Use This Tool