Knowledge structure
Explore images, links, interpretations, relationship types, comments, reactions and authors. View the whole graph or one link.
IMAGES / RELATIONSHIPS / WRITING
Instructions for the current implementation. This document does not read or modify workspace data or API settings. It checks authentication only to enable the design-system link.
Drop images, connect ideas and create a writing guide for AI. There are no sample nodes initially. After sign-in, an empty-canvas hint disappears when a node is created. Saved work loads from the same browser and location.
Start on the canvas| Action | Effect |
|---|---|
| Click a node or link | Select it and open its interpretation editor alongside it. |
| Drag a node | Move the node. |
| Drag empty space / Scroll | Pan / Zoom |
| Click empty space | Clear selection and close the interpretation editor. |
| Shift+click | Adjust multiple selections. Nodes do not move while Shift is held. |
| Fit all | Fit all nodes in view. |
Close writing and ontology panels using their own Close button. Clicking outside or pressing Esc does not close the writing panel.
A thumbnail node appears immediately. PNG, JPEG and WebP images up to 2MB each are supported.
Enter the image source (up to 500 characters) and interpretation (up to 200 characters), then save.
If a source or interpretation is already saved, the button reads Update.
Dropping a new image onto an existing node preserves the original image and creates a new node nearby with a directed link from the original. Fill in the new node first; click the link later to interpret it.
Local mode checks the browser identity; shared mode checks the account ID for node editing and deletion. Older nodes with unknown creators are read-only. Changing a display name does not change ownership.
Select an item and use Delete in the toolbar. Deleting a node also removes its links, while other nodes remain. Local deletions can be undone. Shared deletions require confirmation and cannot be undone. Delete/Backspace shortcuts apply only when the canvas has focus, not while the writing panel is open.
Click the source and target nodes in order. Links are directed.
Define the relationship in the editor beside the link. Add several interpretations, then save them together.
Comments appear in creation order. Agree or veto each comment, then change or remove your reaction if needed.
Deleting a comment also removes its reactions, but not the interpretation or node. To delete the link itself, select it on the canvas and use Delete in the toolbar.
A relationship named Cause is still a proposed interpretation. The ontology does not establish causality as fact.
First, write one sentence explaining which part of A relates to which part of B, and how. Choose the relationship that fits that sentence, then describe its scope and your reasoning in the definition.
In A → B, A is the source node and B is the target. The arrow alone does not imply causation, chronological order or approval. The examples below illustrate ways to read a relationship; its actual meaning must be read together with the participant’s definition.
The Cause and Effect examples both use the direction “A leads to B.” Cause emphasizes A’s contribution; Effect emphasizes the change in B. Selecting Effect does not reverse the arrow. If you mean “A is a result of B,” state this explicitly in the definition, or reverse the link to B → A and review the definition.
| Relationship | When to choose it | A → B example and definition |
|---|---|---|
| Cause | Choose this when interpreting A as causing or contributing to B. | A introduces automation → B has less repetitive work: “Automation is interpreted as contributing to the reduction in repetitive work.” |
| Effect | Choose this to emphasize the change or outcome in B attributed to A. In this example, A is the cause and B is the effect. | A introduces automation → B has greater demand for retraining: “Demand for retraining is interpreted as an outcome of introducing automation.” |
| Premise | Choose this when A is an assumption or condition needed for the argument in B. | A ensures data quality → B uses AI analysis: “The analysis is used on the condition that the data is accurate.” |
| Evidence | Choose this when data, observations or examples in A support a claim in B. Providing evidence differs from claiming a cause. | A user survey → B argues for training: “Difficulties reported in the survey are presented as evidence for the need for training.” |
| Complement | Choose this when A and B fill gaps in each other and are useful together. | A statistics → B interviews: “Statistics show the scale of change, while interviews describe how people experience it.” |
| Contrast | Choose this to highlight differences between A and B using the same comparison criteria. They need not oppose or conflict with each other. | A rapid adoption → B phased adoption: “Compare the pace of adoption and approaches to risk management for the same goal.” |
| Similarity | Choose this to describe a feature or structure shared by A and B. Similarity does not establish the same causes or conclusions. | Case A → case B: “Both cases are similar in that users received insufficient training.” |
| Part and whole | Choose this when one item is a component of another. State which is the part and which is the whole. | A privacy policy → B AI governance: “A is one of the policies that make up B.” |
A premise is a condition for an argument; evidence is material that supports a claim. Contrast compares differences; opposition is a stance against a claim or proposal.
This family is not limited to political parties or political news. Use it to describe how people, groups or positions relate to each other, or how they act together.
| Relationship | When to choose it | A → B example and definition |
|---|---|---|
| Support | Choose this when a person, group or position in A backs a proposal or action in B. | A user group → B proposal to expand training: “A supports putting B’s proposal into practice.” |
| Opposition | Choose this when A objects to or rejects a claim or proposal in B. | Group A → B full rollout proposal: “A opposes B’s rollout proposal because of its cost.” |
| Agree | Choose this when A and B share a position on a particular judgment or claim. Agreement does not necessarily imply practical support. | A expert opinion → B research conclusion: “A agrees with B’s assessment of the risk.” |
| Conflict | Choose this when the goals, interests or demands represented by A and B clash. | A cost-cutting goal → B demand for job security: “The two demands conflict over staffing decisions.” |
| Cooperation | Choose this when A and B combine their roles or resources to pursue a shared goal. | A university → B company: “The university provides teaching and the company provides placements to train people together.” |
| Alternative | Choose this when presenting A as an option to consider instead of B. State what it would replace. | A phased pilot → B immediate full rollout: “A is proposed as an alternative to reduce the cost and risk of B.” |
A useful template is: “The [content] in A has a [relationship] to the [content] in B, because [material or reasoning], within [scope or uncertainty].” Explain how you interpret the connection rather than simply repeating the relationship name.
Example: “Responses in A’s survey reporting insufficient training are treated as evidence for B’s proposal to expand training. The survey covers one organization, so the interpretation is not generalized to the whole industry.”
Even for comparisons such as Similarity, Contrast and Complement, the system does not automatically create a reverse link. State your comparison criteria in the definition. Add a reverse link and interpretation only when you need to record a separate meaning in the other direction.
Support or Agree as a relationship type describes the connection between A and B. Support, Do not support and Undecided on a comment describe a participant’s stance toward that interpretation. Agree or Veto on a comment is a reaction to that comment. Reaction counts do not establish whether an interpretation is true or adopted.
If participants read the same link differently, use Add another interpretation to keep each definition. The AI receives the saved relationship names, definitions, authors and comments together, so clear definitions help it follow the intended meaning.
New canvas saves the current canvas and opens an empty one. Only canvases containing nodes become chips. Repeatedly creating an empty canvas adds no chips. Labels use the first saved timestamp, YYYY-MM-DD HH:mm:ss, and remain stable. Click a chip to return. Deleting a personal canvas permanently removes its local contents after confirmation; removing a shared chip only removes the list entry. Lists are stored per account in this browser. Back up each canvas to move devices.
The top button opens a wide drawer from the right and shifts the canvas left. Closing and reopening preserves input. Close this panel before using backup/restore or creating links.
Start with API settings. Choose Gemini, OpenAI, Claude or a local OpenAI-compatible API. Enter the key and model, then save. Load models or enter an ID manually. Changing connection details cancels an in-flight model request; load the list again. Local APIs also require a server URL.
Save settings again after changing the provider, key or model. API settings remain only in this page’s memory; enter them again after closing the browser or tab, reloading or navigating away. They are excluded from workspace backups. Saving settings does not generate text.
The whole canvas is included, regardless of selection: nodes, link directions, interpretations and comments, including disconnected nodes. Choose genre, purpose, audience, format, length, perspective and principles, then save the guide.
Choose an explanatory article, essay, report, column or news article for nonfiction, or a story or essay for literature. Each genre remembers its selected format. A different format in an older guide remains as saved and is not changed automatically.
Save again after changing materials or settings. AI receives the saved guide text; original images are not sent.
Use Ask AI to write or Generate text explicitly. New generation is blocked until API setup is complete, but saved/shared results remain readable. Opening a tab does not call AI. Requesting text from the guide moves keyboard focus to the result tab.
Cancel while generating. Failed or cancelled requests retain the previous result. Copy the result or download it as TXT.
Usage is billed to your provider account. If direct browser access is blocked, check provider or local-server CORS settings. Review generated text and sources before use.
Review the structure and style rules for the selected format. Seven composition plans distinguish nonfiction essays from literary essays. News articles use a headline, lead and body, with instructions not to invent reporting or quotations absent from the material.
Saving a guide also records its writing-rule version. Review the saved rules and full AI input. Older guides without a version use the current rules; saving them again records the version.
Let AI choose the structure or order nodes manually. Move earlier/later changes presentation order only, not link direction, chronology or causality. Save the guide after changing it.
AI receives the same vocabulary semantics shown in the ontology definitions plus the saved network of nodes, links, interpretations, comments and reactions. Targets and authors are distinguished; uninterpreted links and disconnected materials remain. Sending these rules does not semantically validate the output.
Knowledge structure includes saved guides, material snapshots, actual AI input, generating model and results. Resaving a guide preserves previous result prompts. If an older structured snapshot is missing, only a reference is shown. Unrecorded requesters and paragraph-level evidence are not inferred. These links do not guarantee factual accuracy.
Open Ontology structure to see the actual JSON-LD graph. Select an item to inspect details, identifiers and applicable constraints.
Ontology opens in a right drawer like writing: 75vw on desktop, full width at 800px or below. Only the viewport shifts. Choose Knowledge structure, Canvas connections or Ontology definitions. Writing and ontology open one at a time; writing inputs are retained. Use Close to return.
Explore images, links, interpretations, relationship types, comments, reactions and authors. View the whole graph or one link.
Inspect image-to-image direction and interpretation. Dashed links await interpretation. Click a link label to explore its knowledge structure.
Inspect RDF/OWL classes and properties, SHACL constraints, class counts and property usage.
Physics adjusts spacing automatically. Pause/resume layout, zoom, pan and drag items. Multiple and reverse relations use separate curves. The item selector provides keyboard access. Graph layout does not alter the canvas. JSON-LD export is enabled only for administrators and requires online verification. Backup/restore is separate.
Displaying SHACL constraints is not the same as executing validation. This screen does not show automated reasoning results or replay change history.
Guests start in Sign in & collaborate with Close disabled. After signing in, Close opens access to the canvas. Sign-out reopens this panel without deleting work. Sign out is in the account card. Retained sessions start on the canvas. Personal headings use the login email; shared canvases use their names. Account & workspace contains sign-in/collaboration and backup/restore tabs in a 75vw drawer (full width at 800px). File backups do not require sign-in. Use Close to exit and arrow/Home/End keys to navigate tabs.
| Location | Purpose |
|---|---|
| Backup & restore | Back up and restore images, nodes, relationships, comments, saved guides and results as JSON. |
| JSON-LD export in Ontology structure | Send the knowledge graph to external tools. This is not a workspace restore file. |
| Author settings (beside the workspace name on desktop; in the top menu on smaller screens) | Changes the display name for future contributions. Existing authors and ownership remain unchanged. |
Work saves automatically in the browser. Check the footer status and back up before moving devices or folders. File backup remains available if browser storage fails.
Import replaces the current work; it does not merge. Review node/link counts before confirming. Back up current work and import downloads the existing workspace first. Another browser may not recognize ownership of older nodes.
Register by email and verify the account. Signing in alone does not share your work.
Name and create a shared copy. You own the copied items; previous author labels remain. Existing reactions are marked as unverified history. The original local workspace remains.
Create a code and send it directly to chosen participants. They sign in and join with the code. It expires after 7 days and can be revoked or replaced by the owner.
Item-level changes save automatically to the database and appear for other participants. Check server save/pending/error states in the footer. Incoming changes wait while you are editing.
| Shared data | Private information |
|---|---|
| Images, node positions, sources, interpretations, directed links, definitions, comments, reactions, saved guides and writing results | API provider, model, key, server URL, authentication tokens, current selection and zoom |
A result's model name records provenance. Others cannot use your API configuration. Shared results remain readable without API setup, but new generation requires your own settings.
Node interpretation editing/deletion, link deletion and comment deletion require authorship. Participants can position nodes and add images, links, interpretations and comments. Deleting a node also deletes its links and their contents. Shared deletions require confirmation and cannot be undone.
Concurrent edits to the same item are not silently overwritten. Your changes are preserved with a notice. Back up your edits and load the latest version to compare. Edits to different items can merge. Keystrokes and mouse cursors are not broadcast live.
Unsent edits are kept separately in this browser after connection failure. Retry saving or reopen the same canvas in the same tab/account to recover. Download a backup before fully closing a tab with pending changes. Returning to local work or signing out restores the previous local canvas.
Expand Personal online copies to save/load snapshots. Users can access their own copies. This is not automatic sync; each save creates a new copy. The most recent 50 appear. Ordinary users have no delete button; administrators can hide/delete copies in Content.
On desktop, My canvases sits on the left. A cool-white background from the shared color tokens extends below the list, with a thin right-hand border separating it from the canvas. At widths of 800px or less, the list becomes a horizontal scrolling row above the canvas. Opening a saved canvas fits its nodes to the available workspace.
Administrators see Users and Content after Backup & restore. Inspect online content by user and hide/delete shared canvases, shared results and personal copies. Hiding is reversible; deletion is permanent. Local-only and downloaded copies are outside this management scope.
Open Donate before New canvas for donation information. Only the last three account digits are shown, with no copy function. Donate with Toss opens a transfer screen; desktop shows a QR code. Only Close dismisses the modal. Donations are optional.
The design-system link requires verified administrator access for worflogy@gmail.com. It is disabled for guests and other users. The user guide is public.
Local means stored in this browser. Signing in does not upload work. Save personal online copies explicitly; shared canvases sync changes to the server. Check the footer status.
Administrators can hide, restore, or delete online content. Administrator controls do not include deleting another user’s account or suspending access. Users can delete their own account separately in the Sign in & collaborate tab. Image files that are still in use may remain in storage after content deletion.
The default author is the login email, including on new canvases. Custom names and existing records are retained. Only your own nodes, interpretations and comments display (me); other participants see the stored name. Ownership IDs determine identity and editing rights, not names. Unverified older records do not receive (me).
Use the Korean/English buttons at the top to switch the interface. Unsaved input and canvas data are retained. User contributions and existing AI results are not automatically translated. New AI requests from the English interface ask for English output.
Korean headings use SUITE, English headings use Space Grotesk, and body text and inputs use Pretendard. Double and single quotation marks appear upright, while the original characters you enter, copy, or save remain unchanged.
There is no feature for editing the text of a submitted comment. In a shared canvas, only its author can delete it; deleting a comment also removes its agree and veto reactions. This is separate from editing or deleting a relationship interpretation.
Manage saved interpretations using Edit interpretation and Delete interpretation below the relationship name. Only the author may make changes. Editing retains existing comments, agreements and vetoes; deleting also removes that interpretation’s comments and reactions. The link and other interpretations remain.
Selecting a link shows its source and target images. The link author can reverse the arrow with Reverse direction. Interpretations and comments remain unchanged and should be checked against the new direction. Shared canvases require the database update for direction changes.
For deletions and important changes, an in-app dialog explains the target and impact. Cancel receives initial focus, and action buttons use explicit labels such as Delete or Switch. Close and Cancel do not perform the action. Clicking outside or pressing Escape does not dismiss the dialog. Completion notices use toasts, and input errors appear near the affected field.
Writing materials are summarized at the top with counts of images, links, interpretations and comments. Expand images to review interpretations and sources, or links to review their source, target, interpretations and comments. All canvas materials are included regardless of which lists are expanded. Original image files are not sent to the AI.
The AI guide summarizes the saved purpose, audience, format, perspective, narrative order and writing principles. Expand View full AI instructions to inspect the actual document. Copy, download and AI requests always use the full instructions, including the ontology and canvas materials.
Node interpretation presents the image and title, source, interpretation field and save action in order. Relationship interpretation shows the direction and interpretation list, with the selected definition, author, edit and delete actions grouped in one card. Expand Leave a comment to write an opinion; existing comments are shown oldest first with agree and veto actions. Existing author permissions still apply.
References are resolved against the material sent to the AI for that result, rather than the current canvas. If the AI omitted only id- and the guide, entity kind and complete UUID match exactly, the reference is linked to that material. Bracketed references containing id- followed by a complete UUID or its first eight characters are also linked when exactly one candidate exists in the historical material. When a node, link, interpretation or opinion kind is specified, matching is limited to that kind. References with no candidate or multiple candidates remain unresolved.
If a reference remains unresolved, expand Original identifier and View original AI response to inspect it. A reference number does not guarantee a source match or fact checking. Copied and downloaded results also retain unresolved references.
Identifier recovery applies only to the readable presentation, including previously saved results. The original AI response and input material remain unchanged.
Writing results display headings, paragraphs, emphasis and lists. Long identifiers become numbered references, with saved node and relationship interpretations listed below the text. Identifiers without matching saved materials are marked as unresolved. Reference numbers do not indicate fact-checking.
Copy and download provide readable text and references. View original AI response shows the exact response, including identifiers and Markdown. Stored responses and AI input materials remain unchanged.
In the same browser, reloading or returning from the user guide or design system reopens the last canvas used by your signed-in account. Shared canvases reopen after server access is checked. If a shared canvas cannot be opened, your personal canvas is shown with a notice. Choosing Return to local work makes that personal canvas your next starting point. The last selection is not synced across devices.
Choose Continue with Google, select your Google account, and return to the canvas. To continue your existing work, use the Google account with the email you used to sign up. If Google setup is not complete, you can still sign in with email. Google sign-in is supported on the published site or localhost, not when opening the HTML file directly.
In Account & workspace management, open Delete account on the Sign in & collaborate tab. Review what will be deleted, enter your login email and confirm permanent deletion. Back up any work you need first. Local work and downloaded files remain.
The AI provider, model, API key and server address remain only in the current page’s memory. Enter them again after closing the browser or tab, reloading, or navigating away. Previously persisted API settings are also removed. Your canvas, saved writing guide and generated text are retained.
The user guide and design system share the same header, contents navigation, and content layout. The return-to-canvas link, document links, and language controls stay visible while scrolling. Contents links position headings below the sticky header.
The footer’s terms of service and privacy policy open in a modal, including when signed out. This works consistently on the canvas, user guide, and design system pages. The document scrolls beneath a header with its title and Close button.
Close the modal using its Close button. Clicking outside or pressing Escape does not dismiss it. Closing returns keyboard focus to the link that opened it. Opening a modal does not navigate away, so the current canvas and API settings being entered remain in place.
On English screens, the legal documents remain in Korean, and their links are marked “Korean.” Changing the interface language does not provide an English translation of the legal documents.