# Intelligent Summaries at the Edge

> Process documents, conversations, and data streams with AI-powered summarization. Automatic chunking, semantic compression, and multi-format support with edge caching.

Canonical: https://summarize.do/

## In code

```ts
import { Summarize } from 'summarize.do'

// Summarize documents at the edge
const summary = await Summarize.document({
  content: longArticle,
  length: 'brief',
  format: 'bullets'
})

// Real-time conversation summaries
on.Chat.Message.sent($ => {
  if ($.conversation.length % 10 === 0) {
    const recap = Summarize.conversation({ messages: $.conversation })
    $.Chat.send.Summary({ recap, threadId: $.threadId })
  }
})
```

## Frequently asked questions

### Why use summarize.do instead of calling AI APIs directly?

Summarize.do runs at the edge with automatic caching (90% cost reduction), intelligent chunking for large documents, and multi-model routing for optimal quality and cost. Handles rate limits, retries, and fallbacks automatically. Pre-optimized prompts for different content types (articles, code, conversations, data) deliver better results than generic API calls. Edge caching means identical content returns summaries in <10ms.

### How do you handle documents larger than model context windows?

Automatic intelligent chunking with semantic boundary detection. Documents are split at natural breakpoints (paragraphs, sections) while preserving context. Map-reduce summarization aggregates chunk summaries into coherent final output. Handles documents up to 1M tokens with maintained coherence. Hierarchical summarization for extremely large corpora (millions of documents).

### What formats and languages are supported?

All major document formats: PDF, DOCX, HTML, Markdown, plain text, and structured data (JSON, CSV, XML). Code summarization for 50+ programming languages. Audio transcription + summarization for meetings and podcasts. Supports 100+ languages with automatic language detection. Preserves formatting, links, and metadata through summarization process.

### Can I customize summarization style and focus?

Yes. Configure output length (brief/medium/detailed), format (bullets/paragraphs/executive), and focus areas (technical details, business implications, action items). Custom prompts and examples for domain-specific summarization. Fine-tune models on your content for improved quality. A/B test different approaches and track which summaries users engage with most.

### How reliable and accurate are the summaries?

Multi-model ensemble with automatic quality scoring and confidence intervals. Low-confidence summaries are automatically regenerated with different approaches. Citation tracking ensures summary claims link to source material. Human feedback loop improves quality over time. Battle-tested across billions of documents with 95%+ user satisfaction scores. Automatic factuality checking flags potential hallucinations.

## Part of Platform.do

summarize.do is one of the primitives of Platform.do, in the Agents layer (reason, research and act). Every .do shares one grammar, Business-as-Code, and one SDK.

- What is Business-as-Code? https://platform.do/business-as-code
- Platform.do: https://platform.do/
- Docs: https://docs.platform.do