Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarTechCrunch→ original

Former Meta Director Launches Ndoc, an AI Alternative to Office and Google Workspace

Rohit Patel, former director at Meta, spent eight years at the company before launching Ndoc, an office suite where AI assembles spreadsheets and presentations on its own. Everything lives in a single document with three tabs: a presentation, a narrative with conclusions and sources, and data with Python scripts. Anyone can open the code and check where a formula came from. For now it is an alpha built by a team of three, but the verifiability of AI output is a lesson for any enterprise product.

AI-processed from TechCrunch; edited by Hamidun News
Former Meta Director Launches Ndoc, an AI Alternative to Office and Google Workspace
Source: TechCrunch. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTechCrunch ↗
Editor-in-Chief
Key metric #1The .ndoc.html format will open in any browser, and the team promises to make it open.
Key metric #2Ndoc runs on a mix of open models, so it costs less to run than frontier AI.
Key metric #3The team has three people and no investment; users pay for tokens and usage.

Over the past two years, major AI labs have been embedding neural networks into the spreadsheets and documents people already use: the model fills in cells, writes paragraphs, and draws slides. Rohit Patel, former director of Meta Superintelligence Labs, considers this approach a dead end and left the company to build an office suite from scratch. His startup, Ndoc, aims to take the place of Microsoft Office and Google Workspace, but with one distinguishing feature: everything the AI does can be verified.

One document instead of a chain of six tools

Patel calls the problem "ChatGPT copy-paste." An analyst pulls data out of Excel, runs it through a chatbot, pastes the result into Word, draws charts in Tableau, and assembles slides in PowerPoint. Ndoc brings all of this into one workspace with three tabs.

The Stage tab is the showcase: presentations, dashboards, simulations, and mini-apps with key findings. The Narrative tab holds the text of the analysis with numbered sources, generated charts, and an explanation of why the numbers are what they are. The Model tab contains the raw data and the Python scripts the whole thing is built on.

Here is how it works: the user uploads a dataset and asks for an analysis and written conclusions. They can set the document's context, such as audience and tone, and Ndoc keeps memory at four levels: the person, the specific document, the team, and the company. Colleagues can open the result even if they are not registered in Ndoc.

Verifiability as the main argument

At Meta, Patel worked on reinforcement learning (where an agent learns by trial and error) and on assessing whether AI answers can be trusted. That shaped the design. If a neural network builds tables and slides that companies use to make decisions, someone must be able to check its work.

For that reason, the scripts are not hidden. Even without knowing Python, a user can ask the built-in chat, "What does this script do? Which line contains my formula?" and confirm the calculation is correct. In Patel's words, "there's nothing hidden behind the curtain." For financial and analytical work, where a mistake in one cell costs money, this is a stronger argument than one more percentage point of model accuracy.

The second bet is the open file format. A document is exported as a single .ndoc.html file that opens in any browser. Patel wants to publish the specification so third-party developers can build compatible applications and users are not locked into one ecosystem, as they have been for years with Microsoft and Google. By his account, this was partly why he left: at Meta, agreeing on an open format would have required the opinions of "hundreds of people."

Economics, competitors, and risks

Ndoc is built and fine-tuned on a mix of open models rather than frontier models. Patel claims this is significantly cheaper to operate, although there is no independent confirmation. Interestingly, early test users are not asking about this: they are pleased that the whole process, from data to conclusion, is gathered in one place.

Right now Ndoc is running a corporate pilot and opening an alpha to a limited number of clients from a waitlist. Registration is free; users pay for usage and tokens. A beta is promised within the next few weeks, with a full launch next year. The team has three people and no external funding: Patel wants to first prove that people like the product and only then raise money.

The risks are obvious. OpenAI, Anthropic, Google, Meta, and others have already released enterprise tools that plug into Google Workspace and Microsoft 365. A month ago OpenAI launched Space, a shared workspace inside ChatGPT that is a direct competitor to Ndoc. Changing the habits of entire companies is hard, so Patel built connectors for importing data from existing services.

For developers and product managers, there are two practical takeaways. First, in AI tools for work, transparency of results can be a selling feature rather than a technical extra. Second, an open document format is a way to compete with the giants without their distribution. Whether this works for a three-person team will be shown by the beta and the first enterprise customers. Patel himself operates on a simple rule: execution is cheap today, so it is better to launch an imperfect plan than to wait for a perfect one that someone else implements first.

Story

Key Questions Answered

What are the key details on One document instead of a chain of six tools?

Patel calls the problem "ChatGPT copy-paste." An analyst pulls data out of Excel, runs it through a chatbot, pastes the result into Word, draws charts in Tableau, and assembles slides in PowerPoint. Ndoc brings all of this into one workspace with three tabs. The Stage tab is the showcase: presentations, dashboards, simulations, and mini-apps with key findings. The Narrative tab holds the text of the analysis with numbered sources, generated charts, and an explanation of why the numbers are what they are.

What are the key details on Verifiability as the main argument?

At Meta, Patel worked on reinforcement learning (where an agent learns by trial and error) and on assessing whether AI answers can be trusted. That shaped the design. If a neural network builds tables and slides that companies use to make decisions, someone must be able to check its work. For that reason, the scripts are not hidden. Even without knowing Python, a user can ask the built-in chat, "What does this script do? Which line contains my formula?" and confirm the calculation is correct.

What are the key details on Economics, competitors, and risks?

Ndoc is built and fine-tuned on a mix of open models rather than frontier models. Patel claims this is significantly cheaper to operate, although there is no independent confirmation. Interestingly, early test users are not asking about this: they are pleased that the whole process, from data to conclusion, is gathered in one place. Right now Ndoc is running a corporate pilot and opening an alpha to a limited number of clients from a waitlist. Registration is free; users pay for usage and tokens.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…