docAnalyzer

Reuse outputs across turns

Files made in earlier turns are referable later: convert, bundle, or feed them back into composition.

A docAnalyzer chat doesn't throw away its work product after each turn. Every file the chat produces stays addressable for the rest of the session. Later turns can convert it, bundle it with other outputs, or feed it back as input to a new composition, without regenerating anything.

This is what makes a chat a workspace rather than a single-turn Q&A.

More than one artifact per turn

A turn isn't limited to one output. Ask for a contract analysis and you may get back the analysis, the extracted-field spreadsheet, and a comparison chart in the same turn. The chat produces what the work needs, not just what you literally asked for. The reuse patterns below operate on those artifacts later; this is what fills the working set in the first place.

Refer to a previous artifact naturally

Once an artifact shows up as a chip in the chat, you can refer to it in any later turn: by name, by what it is, or by ordinal:

  • "Convert the spreadsheet to PDF."
  • "Bundle the summary and the chart into a ZIP."
  • "Use the extraction as the source for the spreadsheet you're about to make."
  • "Re-style the brief as a one-page version."

The chat resolves these natural references against the artifacts in the session.

Three reuse patterns

Convert

Take an output and produce a different format from it.

"Convert the spreadsheet to a PDF I can email."

The chat calls the matching converter (Pandoc, LibreOffice, or Calibre depending on the source/target pair) and produces a new artifact. The original stays untouched.

See What you can generate for the full convert matrix.

Bundle

Combine multiple outputs into a single archive.

"Bundle the extraction spreadsheet, the summary PDF, and the chart into a ZIP I can share."

The chat invokes the bundle tool and produces a ZIP that preserves the original filenames.

Feed back as input

The cleanest reuse: pass an output back as material for the next composition.

"Use the data in the spreadsheet to make a bar chart by region."

The chat reads the existing artifact, doesn't re-emit the underlying data, and produces a new chart grounded in it. This is how you build up complex final outputs over a few turns instead of cramming everything into one mega-prompt.

Worked example: building a briefing pack

A four-turn sequence that produces a complete briefing pack:

  1. Turn 1: "Extract these fields from each contract: counterparty, term, renewal, governing law." → XLSX artifact.
  2. Turn 2: "Summarize each contract in a paragraph." → ZIP of per-contract markdown summaries.
  3. Turn 3: "Make a chart of contract terms by counterparty, using the extraction." → SVG bar chart.
  4. Turn 4: "Bundle the extraction, the summaries, and the chart into a ZIP called Q3-contract-pack." → ZIP artifact.

Download the bundle. The whole pack is one file.

What doesn't carry over

  • Artifacts are session-scoped. They don't survive across separate chat sessions or workspace switches. Artifacts from one chat are not accessible from a different chat.
  • The 12-output cap still applies. If you're at 12 and produce more, the oldest gets evicted. Save anything you need to keep before the cap closes in.
  • Cross-session reuse isn't supported. If you want to use yesterday's chart in a new chat, download it from the old session and re-upload it as a document (or paste it as a note).

For API consumers

Each artifact carries an OUTPUT!N reference (OUTPUT!1, OUTPUT!2, …) that appears in answer bodies returned by the API. When you build on top of the API, reference artifacts by this token in follow-up prompts; the chat surface in docAnalyzer hides the token and resolves natural references for you.

What's next

Was this helpful?