Solutions · Document Intelligence · Summarization

Hundreds of pages. One paragraph.

The Summarizer engine handles documents of any size, including those larger than the model's context window. Recursive summarisation breaks long inputs into chunks, summarises each, then summarises the summaries. Works on PDFs, DOCX, images, scanned docs via OCR, and plain text. Configurable length, intent, and auto-title.

Start building free API reference

Recursive overflow 3 intents Auto-title

Recursive summarisation

Documents larger than the context window are chunked, summarised, then re-summarised. Lossless of intent.

Three intents

Executive, bullet, narrative. Pick the shape your reader expects.

Vision aware

Summarise scans, screenshots, and image-only pages via VLM-backed OCR.

What the engine handles

Summarisation at any scale.

Any size input

Recursive overflow strategy splits long documents into chunks, summarises each, then summarises the summaries. No 200-page limit.

Three styles

SummaryStyle: Executive (one paragraph), BulletList (key points), Narrative (flowing prose).

Configurable length

SummaryLength: Short, Medium, Long. Or set an explicit token target.

Auto-title

Generate a concise title alongside the summary. Useful for indexing libraries of unnamed scans and reports.

Streaming output

Token-by-token streaming via SummarizeAsync. Render in real time as the model writes.

Vision-grounded

Pass a vision-language model and the engine pulls text from images, screenshots, or scans automatically.

Three lines

Document to summary.

Summarize one document with an executive style, a short length, and recursive overflow handling.

SummariseReport.cs

using LMKit.Model;
using LMKit.TextGeneration;

var model = LM.LoadFromModelID("qwen3.5:4b");
var summ  = new Summarizer(model)
{
    Style  = SummarizerStyle.Executive,
    Length = SummaryLength.Short,
    OverflowStrategy = SummarizerOverflowStrategy.RecursiveSummarize,
};

// Works on PDFs, DOCX, plain text, images. Hands-off.
var result = await summ.SummarizeAsync(@"C:\reports\Q3-annual.pdf");

Console.WriteLine(result.Title);
Console.WriteLine(result.Summary);

Summarize each weekly report into bullets, then summarize the summaries into an annual narrative.

BatchArchive.cs

// Summarise a year of weekly reports into one digest.
var reports = Directory.EnumerateFiles(@"C:\weekly-reports", "*.pdf");
var bullets = new List<string>();

foreach (var r in reports)
{
    var s = await summ.SummarizeAsync(r, intent:
        "One bullet per major decision or risk surfaced this week.");
    bullets.Add(s.Summary);
}

// Now summarise the summaries.
var annual = await summ.SummarizeAsync(string.Join("\n\n", bullets),
    style: SummarizerStyle.Narrative,
    length: SummaryLength.Long);

Demos and guides

Working references.

DemoDocument summarizer GuideBuild a summarisation pipeline GuideSummarise documents and text

Related capabilities

Summarisation plus the rest.

Content summarisation (text)

The same engine for plain-text inputs without document parsing.

Text summarisation page

Document Intelligence chat

For interactive Q&A over the same documents you summarise. Different surface, same corpus.

Chat with PDF page

OCR

Scans and image-only PDFs run through OCR transparently before summarisation.

OCR page

Document monitoring agent

Watch a folder, summarise on arrival, route to the right channel. End-to-end automation.

How-to guide

Demos & docs

Build it. Read it. Try it.

Working console demos on GitHub, step-by-step how-to guides on the docs site, and the API reference for the classes used on this page.

Demo

Any size. One summary.

Get Community Edition Download