Skip to content

DOCUMENT INTELLIGENCE PIPELINE

Turn complex documents into reliable knowledge.

DeckOps turns PDF, Office, and iWork documents into reliable, usable knowledge—preserving reading order and critical structure so AI can retrieve, understand, and operate on complex documents.

Future-state product contract. The current public legacy repository does not prove the format-specific IR pipeline and has no effective repository license.

Evidence boundary
StatusBoundary
CurrentNo format-specific IR pipeline is proven by the public legacy repository.
Release-scopedFuture-state product contract; verbs stay release-scoped until fixtures prove them.
PlannedFormat-specific parse route, IR, and same-format export direction.
Future-state contractGROUNDED EXPLANATORY ARTIFACT
  1. 01DeckProbe result
  2. 02format-specific parse route
  3. 03IR
  4. 04Markdown / RAG / extract / modify / same-format export direction
Why structure matters

The document intelligence pipeline for agents.

Complex documents carry meaning through reading order, regions, relationships, and visual hierarchy. Loose text alone cannot preserve that context.

  1. 01

    Preserve reading order and critical structure before retrieval or operation.

  2. 02

    Format-specific representation

  3. 03

    Reliable knowledge

GROUNDED EXPLANATORY ARTIFACT

Route-to-IR contract

Route the document before downstream work.

This is the future-state Product contract, not a released universal pipeline.

  1. 01Phase 1

    DeckProbe result

  2. 02Phase 2

    format-specific parse route

  3. 03Phase 3

    native IR

  4. 04Phase 4

    Markdown / RAG / extract / modify / same-format export direction

GROUNDED EXPLANATORY ARTIFACT

Direction versus current source

Separate the Product contract from the legacy repository.

Future-state Product direction

Targeted DeckProbe routing, format-specific parsing, a native IR, preserved structure and visual reading order, and downstream document work remain the intended contract.

Current legacy repository context

The public repository documents TypeScript and Go SDKs, a Node CLI, authentication, uploads, cloud tasks, and task waiting. It does not prove the new IR contract and has no effective repository license.

View the current repository context
Project relationships

How the projects work together

DeckProbe

An independent packageable project whose targeted result routes DeckOps parsing.

Parse

An internal DeckOps stage, not a Product or website destination.

DeckRender

An independent output project that DeckOps may use; rendered output does not prove semantic QA.

DeckUse and DeckHTML

Distinct Products: a lossless Office-file runtime and a lifecycle-qualified web-content bridge.

Evidence needs

What would unlock execution

  • A release or commit binding for the new Product contract.
  • A per-format parser, IR, and operation matrix.
  • Reproducible fixtures and maintained interfaces.
  • An effective repository license and documented data-handling boundaries.
FAQ

Frequently asked questions

Answers about DeckFlow Products, open source, formats, and pricing.

DeckFlow is open-source document infrastructure for AI agents. It gives agents three Product paths—DeckUse, DeckOps, and DeckHTML—plus ready-to-use Solutions for presentation and document workflows.

Choose by the lifetime of the work. DeckUse keeps working on the same document with persistent state. DeckOps finishes one bounded document task. DeckHTML moves content between Office files and the web.

Yes. DeckUse is AGPL-3.0. DeckHTML, DeckProbe, and DeckRender are MIT. DeckOps does not yet have an effective repository license on the public legacy repository.

DeckUse current reviewed behavior is scoped to existing PPTX. DeckHTML current path converts HTML, URL, stdin, or SVG to PPTX or PNG. DeckOps remains a future-state product contract until a release proves format-specific IR claims.

DeckFlow uses one-time Credit packs. Credits pay for supported DeckUse, DeckOps, and DeckHTML tasks as each Product's current release allows. See the pricing page for pack sizes.

No. Solutions are job-led workflows—such as translate, generate, and revamp—that run on the Product paths. They are not a separate Product layer.

Next steps

Continue with architecture and status paths.

Future-state product contract. The current public legacy repository does not prove the format-specific IR pipeline and has no effective repository license.

Related resources