Nomic

Nomic AI develops tools that make AI systems and their data more accessible and explainable. Their primary products, Atlas and GPT4All, enable users to interact with massive datasets and run AI models on consumer-grade hardware, respectively.

Key features:

- Data Structuring and Visualization: Instantly structure, visualize, and derive insights from millions of unstructured data points across various modalities (text, image, audio).
- Collaboration and Integration: Enables collaboration for technical and non-technical teams and integrates into production workflows with cloud and on-premises hosting options.
- Advanced Data Operations: Offers vector search, topic modeling, semantic deduplication, embeddings, and tools for data refinement and cleaning.
- User Interaction: Allows users to interact with data semantically from their web browser or through the Atlas API, supporting the creation of shareable maps with or without coding knowledge.
- Model Deployment: Enables running open-source AI models on consumer-grade CPUs and GPUs from NVIDIA, AMD, Apple, and Intel.
- Customization and Inference: Allows for model customization with retrieval-augmented generation and runs inference without requiring a GPU or internet connection.
- Local and Secure AI: Focuses on privacy-oriented software for running large language models locally on various hardware configurations.

How it works:

1. Users explore, label, search, and share massive datasets directly from their web browser.
2. The platform supports various data modalities, including text, image, audio, and video.
3. Users can download and install GPT4All models, which range from 3GB to 8GB in size.
4. The software ecosystem supports running these models locally on consumer-grade hardware.

Integrations:

Hugging Face, MongoDB, and others. Supports embedding models and vector search capabilities.

Use of AI:

Atlas utilizes advanced embedding models like Nomic Embed, an open-source, fully auditable text embedding model with an 8192 context size. GPT4All is built on top of models like GPT-J and other large language models, trained using large-scale data distillation from models like GPT-3.5-Turbo.

Target users:

- Data scientists, machine learning engineers, and researchers who need to interact with and derive insights from large unstructured datasets.
- Organizations looking to deploy AI models locally for privacy and security reasons.

How to access:

Atlas is available as a web app, API, and Python client. GPT4All can be downloaded for Windows, macOS, and Linux operating systems.

Open Source Status:

Both Atlas and GPT4All are open-source projects, with repositories available on GitHub.