The New York lab making AI - and the massive, messy datasets that power it - something anyone can see, run, and trust.
Nomic AI, New York City. The company behind Atlas, Nomic Embed, and GPT4All - three open tools that pulled AI's inner workings out of closed labs and into the browser.
Most AI arrives as a black box: a model you cannot inspect, trained on data you cannot see, running on servers you do not control. Nomic AI was built on the opposite bet - that transparency is not a liability but a strategy.
Founded in 2022 in a Manhattan office on East 20th Street, Nomic AI is the work of three founders - Andriy Mulyar, Brandon Duderstadt, and Ben Schmidt - who set out with an unusually plain mission: improve the explainability and accessibility of AI. They treated AI as a once-in-a-century technology that everyone, not just a handful of well-funded labs, should be able to participate in and benefit from. That conviction shaped an unusual product line and, later, an unusual pivot.
The company's first product, Atlas, is a data engine with a scalable embedding-space explorer. In plain terms: it takes a pile of unstructured data - hundreds of points or tens of millions - and turns it into an interactive map you can pan, zoom, search, and clean, right inside a browser tab. Text, images, audio, video: all of it can be placed into the same visual space of meaning. For a machine-learning engineer staring at a dataset they cannot describe, Atlas answers a deceptively hard question - what is actually in here?
If Atlas made data visible, GPT4All made models portable. Released in 2023, GPT4All is an open-source ecosystem for running large language models locally, on ordinary consumer hardware. No cloud account, no data leaving the machine. It became one of the most downloaded local-LLM projects on the internet, and it planted a flag that would define Nomic's identity: private, inspectable, on-device AI.
Teams sit on huge, unlabeled datasets with no way to inspect them. Atlas turns that fog into a navigable map - spotting duplicates, outliers, and mislabeled clusters at a glance.
Sending sensitive documents to a third-party API is a non-starter for many. GPT4All runs models locally, so data never leaves the device - private document chat, fully offline.
Keyword search misses intent. Nomic Embed converts text, code, and images into vectors so systems can retrieve by meaning - the quiet engine behind modern RAG.
Closed models hide their training. Nomic ships open weights, open code, and open data so researchers can reproduce and verify results themselves.
General models fumble on drawings and specs. Nomic's AEC platform is tuned for the built world - drawing review, code compliance, submittals, and RFIs.
Most visualization tools choke on real datasets. Atlas is built to render and search datasets spanning tens of millions of points without leaving the browser.
"AI is a once-in-a-century innovation that everyone should be able to participate in and benefit from - so the models and the data behind them should be open to inspect."- Nomic AI, on its mission
A browser-based data engine and embedding-space explorer. Visualize, curate, search, and share datasets across text, image, audio, and video - from hundreds to tens of millions of points.
An ecosystem for running LLMs locally on consumer hardware. Private, offline document chat and chatbot deployment - no data leaves the device.
An open, fully reproducible long-context (8,192 token) text embedding model, Apache 2.0 licensed, that outperformed OpenAI's Ada-002 on short and long context tasks.
A multilingual embedder and the first embedding model to use a Mixture-of-Experts architecture, trained on 1.6 billion high-quality data pairs.
State-of-the-art models for visual document retrieval and code search, processing interleaved text, images, and screenshots.
A domain-focused platform applying Nomic's retrieval stack to architecture, engineering, and construction - drawing review, code compliance, submittals, and RFI responses.
Where much of the industry ships closed APIs, Nomic competes on openness, locality, and domain depth. The chart below sketches its relative emphasis - an illustrative read of positioning, not a benchmark.
Against OpenAI and Cohere - whose embedding models are strong but proprietary - Nomic's pitch is that you can read the recipe. Against pure infrastructure players like Pinecone or Weaviate, Nomic owns the model layer as well as the visualization on top. And in the AEC vertical, where incumbents like Autodesk and Trimble own the workflow, Nomic positions itself as an assistive layer for document-heavy tasks rather than a replacement for the specialists who do them.
Nomic gives away model weights, code, and data - which builds credibility, community, and adoption - then earns revenue from its hosted platform. Atlas and the Nomic developer API run on usage and subscription tiers, and an enterprise AEC product adds features like data residency and private deployment. The open releases drive awareness at the top; the platform and vertical product capture value below.
The loudest part of AI is generation - the chatbots. The load-bearing part is often retrieval: turning content into embeddings so systems can find what matters. Nomic sits squarely in that layer, and pairs it with the visualization tools to understand what the data contains. Its 2025 move into construction is a bet that the biggest wins come from taking that stack deep into one under-digitized industry.
Use Atlas and Nomic Embed to explore datasets, build retrieval systems, and reproduce open results.
Individuals and enterprises run GPT4All for local, offline document chat where data cannot leave the device.
Architecture, engineering, and construction teams use the platform for drawing review, compliance, and knowledge search.
Reach: GPT4All and the open embedding models have been downloaded in the millions across GitHub and Hugging Face.
Andriy Mulyar, Brandon Duderstadt, and Ben Schmidt start the company to make AI and its datasets explainable.
A browser-based embedding-space explorer for large, unstructured datasets becomes Nomic's first product.
Nomic's open-source ecosystem for running LLMs locally becomes one of the most popular projects of its kind.
Coatue leads a Series A at roughly a $100M valuation to fund open-source AI for all.
An open, reproducible, long-context embedding model outperforms OpenAI's Ada-002.
A Mixture-of-Experts multilingual embedder ships alongside state-of-the-art multimodal and code models.
Nomic focuses its platform on the built world and expands data residency across AWS regions worldwide.
Nomic publishes a construction AI benchmark and is named a finalist in Trimble's 0-60 Challenge.
CO-FOUNDER & CEO
Leads Nomic's strategy, including the 2025 pivot toward verticalized AI for construction.
CO-FOUNDER & CTO
Drives the technical direction behind Atlas, GPT4All, and the Nomic Embed models.
CO-FOUNDER & VP, INFORMATION DESIGN
Shapes how Nomic visualizes and communicates the structure hidden in large datasets.
Early backing from Contrary Capital, Betaworks Ventures, and SV Angel.
With Contrary Capital, Betaworks Ventures, SV Angel, Story Ventures, and Factorial Capital. Reported at roughly a $100M post-money valuation.
Qualcomm - optimized GPT4All and Nomic embeddings for local inference on Snapdragon X Series devices. Trimble - named a finalist in the 0-60 Challenge 2026.
The name "Nomic" echoes a game whose rules the players can change mid-play - fitting for a company reshaping how AI is built and inspected.
A team of roughly 17 people shipped an embedding model that outperformed one from OpenAI.
GPT4All can run a capable chatbot entirely offline on a laptop, with no data ever leaving the device.
Atlas can place text, images, audio, and video into the same interactive map of meaning.
Nomic pivoted from horizontal open-source tooling toward construction - one of the least digitized sectors in the economy.
Nomic Embed Text v2 was the first embedding model to adopt a Mixture-of-Experts architecture.
Nomic builds tools to make large, unstructured datasets and AI models understandable and usable - notably Atlas for data visualization, the open-source Nomic Embed models, and GPT4All for local LLMs. Since 2025 it applies this stack to the architecture, engineering, and construction industry.
Nomic AI was founded in 2022 in New York by Andriy Mulyar (CEO), Brandon Duderstadt (CTO), and Ben Schmidt (VP of Information Design).
Nomic raised a $17M Series A led by Coatue in 2023, at roughly a $100M post-money valuation, following earlier seed backing from investors including Contrary Capital, Betaworks, and SV Angel.
Yes. Nomic releases many of its models with open weights, training code, and data - Nomic Embed and GPT4All are both distributed under permissive open-source licenses.
Atlas is Nomic's browser-based data engine that lets anyone visualize, curate, search, and share datasets of up to tens of millions of points across text, image, audio, and video.