Skip to content

roadmap.md

Latest commit

 

History

History
86 lines (61 loc) · 6.26 KB

roadmap.md

File metadata and controls

86 lines (61 loc) · 6.26 KB
Raw

Roadmap

v0.1 — shipped

  • pptx pdf key docx html md and URLs → image and PDF
  • ppt → image and video
  • pptx → video
  • Cloud engine over @deckops/sdk
  • Profiles, --json envelope, page selection, zip and stdout output
  • Credentials shared with DeckHTML and DeckOps; guest mode without login

Rendering is the cloud's job

DeckRender is a client. Every conversion runs in the DeckFlow cloud, and that is a deliberate boundary rather than a stage on the way to something else:

  • No local render path. There is no local engine, and one is not planned. The earlier milestone for a LocalEngine — Playwright for HTML, pdfium for PDF — is dropped, which is why the version numbers below shift up by one.
  • A format the cloud cannot convert is not supported. It reports not_implemented and names the missing backend task. The fix is a converter upstream, tracked under Upstream asks.
  • No local logic to widen the matrix. No format is answered with something approximate — an embedded thumbnail, a partial extraction — because that is easier than waiting for the backend. Supported means the real conversion.

RendererOptions.engine still lets a caller substitute their own backend. That is their choice to make; it is not DeckRender shipping one.

Also planned: switch URL input to the backend's html.getByURL once it reaches a published SDK release, for true runtime-DOM capture instead of a local fetch.

Coming soon

Combinations the CLI marks soon and reports as not_implemented. Every one is a missing backend task, and every one is fixed upstream:

Combination Blocked on
.ppt → pdf A task that normalizes the legacy binary file to .pptx before the PDF converter
.xlsx → image, pdf A spreadsheet converter — a workbook needs a real layout pass
.pages → image, pdf A Pages converter
.numbers → image, pdf A Numbers converter
.pdf → video A convertor.pdf2video task
.key → video A convertor.keynote2video task

.pages and .numbers were briefly answered with the first-page preview every iWork document embeds. That was removed: a thumbnail reported as a successful render is worse than a clear "not yet". They come back when the backend can convert them.

v0.2 — video parameters

--fps, --duration and --transition. Blocked upstream: convertor.ppt2video currently accepts no parameters at all (ConvertPptToVideoParams = Record<never, never>), so the flags fail today rather than silently rendering backend defaults.

v0.3 — throughput

  • Batch rendering across many inputs
  • --watch
  • Render cache keyed on input hash plus parameter hash

v1.0 — frozen contracts

The CLI output contract — the --json envelope, error codes, exit codes and artifact layout — becomes stable under semver.

The agent-oriented view — images plus element bounding boxes — is composed at the DeckFlow level from DeckProbe's structural output and DeckRender's images. DeckRender will not parse documents to produce it; that would cross the boundary this tool is built around. Its only obligation is that page numbers in --json match DeckProbe's.

Upstream asks for DeckOps

Several limits here are not DeckRender's to fix:

Ask Unlocks
Fix @deckops/sdk package.json: types points at ./src/index.ts, which files does not publish Removes the tsconfig paths workaround every TypeScript consumer currently needs
Publish html.getByURL in a released SDK Runtime-DOM capture for URL input
Parameters for convertor.ppt2video (fps, durationPerSlide, transition) v0.2
Resolution and format parameters for convertor.keynote2image --width and --image-format jpg on Keynote input
Page ranges on conversion tasks Makes --pages save compute and cost, not just bandwidth
Task chaining by upstream task id Removes the download/upload round-trip in every derived route
A convertor.html2pdf task Removes the HTML→PDF fidelity caveat
Pages and Numbers converters .pages and .numbers input
convertor.pdf2video and convertor.keynote2video tasks .pdf and .key to video

Out of scope

DeckRender renders. It does not parse content, extract text, understand structure, or edit files. Those belong to DeckProbe, DeckUse and DeckOps.

It also does not render locally. Bundling a browser, a PDF rasterizer or a format-specific extractor would make this a second renderer competing with the backend, and every format it could reach that way would be one the cloud never got asked to support.

Those boundaries are what keep the CLI small: one command, a document in, pixels out.