Sources & content

Files & source processing

Upload PDFs, Office files, and images—or paste links and ask for web research in chat. The Agent extracts key points and data into your brief for outlines, charts, and slide generation.

How it works

Your sources are the input layer for a deck: you upload originals, the Agent parses them into readable summaries and structured data in the brief. Outlines, charts, and slides then reference this organized content instead of re-reading raw files every time.

The Brief tab splits into **Sources** and **Project materials**: Sources holds originals and extract summaries under `_source/`; Project materials holds embeddable figures, illustrations, analysis PNGs, research notes, and more. When parsing PDFs, the Sources preview shows ingest status and can jump to matching embedded figures.

Brief tab and extracted summaries · coming soon

Supported formats

Documents: PDF, Word (.docx), PowerPoint (.pptx), and other common types.
Spreadsheets: Excel (.xlsx), CSV, and similar—extracted as structured data.
Images: PNG, JPEG, etc., as visual references or assets.
Text: plain text, Markdown, and more.

Each file has a size limit; unsupported types show a clear error. Quality varies with file complexity; scanned PDFs may take longer or parse with limits.

Where to upload

Brief tab: drag or pick files to manage project sources in one place.
Chat attachments: add or drop files with your message to the Agent.
New project: attach initial files on /new before entering the workspace.

Links & research

Besides uploading files, you can bring web sources and public information into the project from chat:

Paste a URL: give an article or report link—the Agent reads the page and files content into the brief.
Web research: when you do not have a file yet, ask the Agent to search public sources on a topic and organize highlights into the brief.
Known URL → paste the link; still need sources → ask for research.

Parse a web page

Read https://example.com/report and extract the key numbers and takeaways into brief.

Web research

Research public info on the 2024 EV market, organize highlights into brief, and cite sources.

After upload

The Agent extracts narrative content and summaries—review them under **Sources**.
Excel / CSV become structured tables—the same numbers feed editable charts and analysis figure PNGs.
PDFs can yield embedded figures (Fig) under **Project materials › Embedded**—visual assets from the source, not reverse-engineered chart data.
Heavy stats (heatmaps, regression, etc.) land on slides and also appear under **Project materials › Analysis**.
In chat: “Organize a brief from this PDF” or “Chart the attachment data on slide 4.”
For files already in the project, @ reference is more efficient than re-uploading.

Example prompts

Parse into brief

Parse the PDF I just uploaded—extract section highlights and key data into a brief.

Attach files in the message, or @ an uploaded asset.

Extract a section only

From this 80-page report, extract only the “Q3 financial performance” sections—ignore the rest.

Narrow scope speeds processing and reduces noise.

Spreadsheet → chart

@sales.xlsx Use the quarterly data for a revenue bar chart on slide 4.

Confirm before next step

Parse the attachment and show me the brief summary first—I’ll confirm before you draft the outline.

Data & charts

After upload, spreadsheets are extracted into structured data. The Agent then picks one of three common outputs per slide:

Editable charts: bars, lines, pies, funnels for standard KPIs and trends. Tune values in the canvas refine panel; charts stay editable in PPTX export.
Analysis figure PNGs: heatmaps, correlations, distributions, regression, large scatter, and other heavy stats. Python computes from the same spreadsheet; figures appear on slides and under Brief › Project materials › Analysis.
HTML tables: long detail lists without one aggregate takeaway.

Quick guide: weekly revenue, trends, share → editable charts; “category correlation,” “distribution/regression,” insights slides → analysis PNGs; long line-item lists → tables. One project can mix all three (e.g. bar chart on one slide, heatmap on another).

Numbers: the Agent does not invent financial figures when reliable data is missing.
Updates: editable charts → canvas refine; analysis PNGs → update the spreadsheet and ask for a refresh.
vs PDF figures: tabular data → charts or analysis PNGs; embedded PDF Fig → Project materials › Embedded.

Chart editing is limited in image mode. For data-heavy decks, choose code mode when creating the project.

PDF embedded figures

Embedded figures in papers and reports can be batch-extracted into **Project materials › Embedded** for direct use on slides. If the same project also has supplementary spreadsheets, prefer Excel/CSV for editable charts; when you only have PDF figures, treat them as visual references—the Agent will not reconstruct exact data points from the image alone.

Extract PDF figures

Extract embedded figures from pages 3–8 of this paper PDF—I need the original Fig for the methods section.

Table + complex analysis

Parse sales.xlsx: monthly GMV as an editable bar chart, plus a category correlation heatmap as an analysis figure PNG.

Tips

Upload core sources before asking for an outline to avoid back-and-forth.
For large docs, ask to extract only sections relevant to your topic.
Sources stay under your account; share links show read-only slides, not original files.

Brief tab

View and manage uploaded sources in the panel.

Upload attachments

Attach files in chat—limits and workflow.

Charts

Generate charts from data and edit values on canvas.

Agent overview

How the Agent drives intake and generation.