문서지능 · Artifact Intelligence

Give it what to report, to whom, why, plus facts and figures, and you get a document draft ready for approval.
Rules handle table of contents, bullet-point conversion, one-page fitting, and Hangul (HWPX) conversion. People just review for accuracy and point out what needs fixing.

Try it first — no installation — https://artifact-intelligence.app

Demo video

Just input your intent and materials → AI designs the document structure → edit the finished document item by item, down to diagrams and phrasing → download as Hangul (HWPX).

▶ Watch full video (41 seconds) · Try it first without installation — https://artifact-intelligence.app

It changes how you work

Word and Hangul give you a blank screen and you fill it all in. 문서지능 changes the way AI and people work together — AI reads, structures, creates, and checks, while people make judgments. It's like having a practitioner by your side.

• Just provide intent and context. Give it the what, to whom, why, plus facts and figures, and you get a draft ready for approval. No need to predefine format, table of contents, or style.

• It shows you the design first. It shows you the document type and table of contents before drafting, and only writes once you approve. It catches "this should be a results report, not a review report" before you write the content.

• AI creates, people verify. You don't start from a blank page, so it's fast, and rules back the quality, so it stays consistent.

• Output as HTML, final version as Hangul (HWPX). Create living documents with tables, diagrams, and illustrations, then move them straight into Hangul files.

• It gets better the more you use it. Rules learn from where people make corrections. The more practitioner touches accumulate, the more accurate the next document becomes.

Quality is measured in three dimensions: structure, style, and design — information flow and table of contents, bullet-point noun-phrase endings and no translationese, markers and visual hierarchy. These three are aligned with writing rules drawn from over 16,000 real reports, and verification gates block narrative endings, pages over one, and missing figures. If something fails, it fixes itself and resubmits.

Related MCP server: mcp-server-docx

Six document types for approval

Each document has a defined format, so the same content renders identically in any environment, down to the fonts.

Your documents never leave this computer

From assessment, drafting, assembly, and checking to Hangul conversion, everything involved in creating documents is processed on your installed computer. Neither the materials you provide, the drafts you create, nor the finished documents are uploaded to a server. The only thing fetched from the server is writing rule fragments, and even those are fetched on demand, only as much as needed for the task.

Installation

Claude Code — via the marketplace:

/plugin marketplace add Kminer2053/Artifact-Intelligence-public
/plugin install artifact-intelligence@artifact-intelligence

Codex — via the plugin marketplace (no clone needed):

codex plugin marketplace add Kminer2053/Artifact-Intelligence-public
codex plugin add artifact-intelligence@artifact-intelligence

Requires Python 3.10+. To use Hangul (HWPX), uploads, and rule lookup, run bash bin/bootstrap.sh once from the installed plugin folder (check the path with codex plugin list).

Cursor — connect via MCP server + rules:

git clone https://github.com/Kminer2053/Artifact-Intelligence-public
cd Artifact-Intelligence-public/artifact-intelligence
bash bin/bootstrap.sh
cp .cursor/rules/artifact-intelligence.mdc ~/.cursor/rules/   # 작업 안내 규칙(선택)

Then add the following to mcpServers in ~/.cursor/mcp.json and restart Cursor (<경로> is the absolute path of the folder above):

"artifact-intelligence": { "command": "<경로>/mcp/run.sh" }

Cursor works through MCP tools (assessment, assembly, gates, export), and placing the rule files above also guides you through the workflow.

Web app — use it right away at https://artifact-intelligence.app without installation.

On installation, you automatically receive a rule lookup token at first launch, and everything needed for Hangul (HWPX) and upload processing is prepared at once. Requires Python 3.10 or later.