Reader

The Reader turns your current selection into a written summary. You pick what goes in, set a length, choose a language, and press Summarize; the panel returns markdown you can read aloud, copy, or download. It is a summarization workbench, not a document viewer.

Where to find it

Open View > Exploration > Reader. Despite what this section of the manual implies, Reader is grouped with the Exploration panels in the app, not the Inspection ones.

The panel needs point metadata to have finished loading; until then you see a spinner.

What it summarizes

The Reader builds one block of text from two sources and sends that to the backend:

• Selected Points. For each included point it sends the title plus whatever it has: date, cluster, layer, partition, keywords, links, and the raw metadata object.
• Chat. The Narrator chat history, if there is any.

Open Include in Summary to change the mix. The Chat checkbox drops the chat transcript; the Selected Points checkbox drops every point; under it, individual points are listed by title with their own checkboxes, three at a time with a “+ N more” link. With nothing selected, the section reads “No Points Selected”; with no chat history, “No messages in chat history”.

Heads up

The Reader does not display the source documents. It lists the titles of the points going into the summary and shows the summary that comes back. To read a point’s own text and metadata, use the Inspector.

Controls

• Summarize / Re-Summarize runs the summary. The label changes once a summary exists.
• The language selector beside the button offers English, French, German, Arabic and Greek. It sets the language the summary is written in, and the voice used for read-aloud.
• Summary Length is a slider from 50 to 1000 tokens, default 100. Under it, Mantis shows the approximate word count and a bucket label: Very short below 100 tokens, Short below 300, Medium below 600, Long above that.

Once a summary exists, four more buttons appear above it:

• Download saves the summary as summary.txt.
• Copy puts it on the clipboard.
• Read sends the summary to the backend text-to-speech endpoint and plays the audio; while it plays the button becomes Stop.
• Clear discards the summary.

The summary itself renders as markdown, collapsed to a preview, with Show More / Show Less once it runs past about 300 characters. Links inside it open a confirmation dialog showing the target URL before anything opens in a new tab.

1. Select points on the map, with the lasso or from a cluster.
2. Open Include in Summary and untick anything you do not want, including the chat transcript.
3. Set Summary Length and, if you want it in another language, pick one from the selector.
4. Press Summarize. The panel shows a spinner while the backend works.
5. Read, listen, or keep it. Use Read for audio, Download for a text file, Copy to paste it elsewhere.

Limits

• The combined text must be at least 100 characters. Below that you get “Content is too short to summarize.”
• With no points included and chat excluded, summarizing alerts “No content to summarize.”
• There is no note taking, highlighting or annotation anywhere in the panel. The only writable controls are the language selector, the length slider, and the inclusion checkboxes.
• Summaries are not saved. Reloading the space, or pressing Clear, loses them. Download or copy anything you want to keep.
• Summarization and read-aloud both call the backend. If NEXT_PUBLIC_BACKEND_URL is unset in a local setup, neither works.

Read nextSummarizerAutomatic, multi-tab analysis of the same selection.