Why we're building ndoc

How it started

A while ago I noticed I'd stopped using standard office tools. I was writing docs in Markdown backed by python analysis, because that's where AI was good. Around me at Meta Superintelligence Labs—and across other labs—colleagues were doing the same, even building React dashboards and sending links instead of slides. People were following the AI to its native languages: React, Markdown, Python.

What makes analytical work and storytelling enjoyable is quickly chase ideas you have, in depth, without getting bogged down by glue work. Yet even with AI, the experience felt clunky and suboptimal: data, analysis, and story were still scattered across separate files, while one-off web pages had no home for the data and reasoning behind them.

I wanted a format native to AI from the ground up—where your quant model, written argument, live slides, dashboards, and mini-apps live in one place. And I wanted it open-source, since we've already spent decades working in proprietary formats. After a graveyard of failed prototypes, ndoc was born with one mission: remove the glue work between tools and deliver an amazing experience for creating and sharing work, so people fall in love with their work again.

Explore the interactive stage

The cost of work about work

People chose their careers out of curiosity and passion for a craft—data science, finance, strategy, marketing, research, engineering. Instead, knowledge workers spend 58% of their workday on "work about work"—acting as the human bridge between AI chats and scattered files:

Re-run the model. Update the sheet. Fix the memo. Rebuild the slide. Check it all still matches. That's glue work. It ate your week.

Combining Asana's 58% benchmark with May 2025 US Bureau of Labor Statistics wage data across 15 knowledge-work occupations (12.97 million US workers), that 23.2 hours a week costs $937 billion a year in wages—an average of $72,267 per worker spent stitching tools together rather than doing the work itself.

Click any linked number or the chart above to open the How was this made? popup and inspect the Python run and dataset behind it.

Why current AI tools fall short

Each individual piece of knowledge work got easier with AI, but holding the pieces together is still your job. Today's knowledge worker uses ChatGPT, Claude, or Copilot at every step—yet each step still produces a separate file across Excel, Word, PowerPoint, Tableau, and standalone mini-app builders:

  1. One project, scattered across files. In every other workflow there are three separate entities: you, an AI, and the files it helps you make. A real project is a quant model, a written memo, a presentation, and an interactive view, but AI can only update one file at a time. When an assumption changes, carrying that change across the sheet, the memo, and the deck is still on you, and the pieces inevitably drift apart.
  2. Numbers lose where they came from. Once a figure lands in a slide or a memo, it turns into dead text. No one can click it to inspect the data and code that produced it, and the chat window where the AI ran the analysis is hard to search and soon expires.
  3. AI pays a tax on old formats. Legacy office files are zipped bundles of deeply nested XML built decades ago for human editing apps, not for AI. Reading and writing them burns massive tokens and model attention fighting formatting tags—making every turn slower, more expensive, and weaker on the actual analysis and writing.
  4. Context lives in a chat, not in the work. The goal, audience, and conventions behind a project stay trapped in one person's chat history. When the work changes hands, that context never travels with it.
  5. Sharing forces a bad choice. You can send a static deck or PDF that your organization knows how to file and permission, but colleagues cannot test a scenario, trace a number, or ask it a question. Or you can send an interactive AI-made web link, which is hard to find later, hard to govern, and still covers only one slice of the project. Either way, every "what if?" comes back to you.

An ndoc that thinks with you: engineered from first principles

ndoc is not one more AI sidebar bolted onto legacy files. It is a fresh take on the office suite that replaces scattered spreadsheets, word docs, slide decks, dashboards, mini-apps, and chat copy-pasting with one unified medium. In ndoc, there are only two things: you (and your colleagues) and ndoc, working together as a team.

1. Three surfaces in one record

Every ndoc brings three connected surfaces together in a single place:

2. AI-native component formats, token efficiency, and visual self-inspection

Every part of an ndoc is stored in the exact open format AI models read and write most fluently:

Because the AI spends zero tokens translating into bloated legacy XML, 100% of its compute and attention goes to the substance of your work—delivering higher intelligence at a fraction of the token cost. And because the AI edits your ndoc with purpose-built tools, it automatically measures slide layouts and inspects rendered screenshots of your stage and narrative, catching and fixing its own visual or layout mistakes before you ever have to ask.

3. Built-in Context & three-level Memory (never by snooping)

ndoc never needs invasive access to your computer, screen, or inbox to understand your work. It sees the bigger picture because the ndoc format is designed to hold the bigger picture:

Seamless flow, effortless sharing, and experiences static files never had

When your idea, data, quant model, written story, and live presentation live in one continuous medium, the glue work disappears. When an assumption changes, the AI re-runs the Python script, updates the dataset and figures, and keeps your stage and narrative grounded in the same source.

Work collaboratively and send the answer, not a dead attachment

What an ndoc looks like in practice

Because one ndoc unites the model, the narrative, and the interactive stage, it adapts to whatever the work requires:

  1. Live Executive Board Deck: A 16:9 presentation with your custom brand theme where slides include live scenario sliders and segment toggles you adjust smoothly mid-meeting, backed by a real dataset and narrative memo.
  2. Financial & Cohort Model: A multi-step quantitative pipeline visualized as a dependency graph in the Model tab, with run-linked numbers in the Narrative that open "How was this made?" lineage on click.
  3. Interactive KPI & Time-Series Dashboard: A multi-chart interactive dashboard combining cross-filtered views with high-frequency time-series exploration over tens of thousands of points.
  4. Team Mini-Applet: An interactive launch tracker, bug tracker, assumption collector, or team calculator that writes directly to its own editable datasets in the Model tab under server-enforced rules.
  5. Research Paper & Literature Review: Publication-grade narrative writing with LaTeX math, numbered citations grounded strictly in verified sources, and reproducible Python figures.

Safe by design, and yours forever