docAnalyzer

How downloads work

Every chat turn can produce downloadable files that stay reusable for the rest of the session.

A chat in docAnalyzer isn't just text. Any turn (in Ask docAnalyzer, in a Focus, or in a workflow run) can produce one or more downloadable files. PDFs, spreadsheets, HTML documents, charts, diagrams, math expressions, ZIPs, code, structured data. Files accumulate over the session and stay reusable for later turns.

What "downloads" means here

"Downloads" is the marketing label for the artifacts a chat produces. Each artifact appears in the answer as an inline chip showing the filename, file size, and a download glyph. Not every artifact gets downloaded; some only exist as input for the next composition step.

There are two ways files get produced:

  • Composed artifacts. The model invokes a compose tool when the answer should be a file: compose_pdf, compose_spreadsheet, compose_chart, etc. The tool returns a structured artifact and the chat surface registers it.
  • Auto-registered text outputs. Structured text outputs (md, json, csv, tsv, html, xml, yaml, svg, txt) are auto-registered as artifacts. Depending on the format and length, an artifact may also render an inline preview in the chat, not only a download chip.

You don't have to pick which path the chat uses. The model decides based on what fits the request.

Produces what the work needs, not just what you asked for

Ask for an analysis of a set of contracts and you may get back the analysis, an extracted-field spreadsheet, and a comparison chart in the same turn, because the work asks for all three. Ask for a lit review and you may get back the review, a summary table of the papers, and a Word version ready to drop into a thesis committee submission.

The chat composes what fits the request and what fits the audience the request implies. If a single text answer is what fits, that's what you get; nothing extra.

Referring to an artifact later

Every artifact the chat produces is addressable for the rest of the session, whether you download it or not. Later in the conversation you can ask the chat to convert, bundle, or reuse a previous artifact ("convert the spreadsheet to PDF", "bundle the extraction and the chart into a ZIP") and it picks up the artifact without regenerating anything. See Reuse outputs across turns for the full pattern.

Under the hood each artifact carries an OUTPUT!N reference (1, 2, 3 … in the order produced). API consumers see this token verbatim in answer bodies; the chat surface resolves natural references like "the spreadsheet" against it.

Session-scoped, not workspace-scoped

Downloads live for the lifetime of the chat session. They are not stored in your workspace. When the session ends, the files are gone.

If you want to keep a download:

  • Download it locally before closing the session.
  • Save the answer as a Note: captures the answer text (with citations), but not the artifact itself.

This is a deliberate design: downloads are working output, notes are durable output. See Save and export your work.

Cap

A single chat session can hold up to 12 outputs. Once the cap is hit, new outputs evict the oldest. This is rarely a constraint; most sessions produce a handful of files.

What's next

Was this helpful?