# Voxel51

> Multimodal data platform for searching, visualizing, curating, annotating and evaluating data behind physical AI and computer vision models.

Voxel51 builds a multimodal data platform, based on the open-source FiftyOne tool, for teams working on physical AI and computer vision. It lets users search, visualize, curate and label data such as images, video, 3D, lidar and radar, and its site says Voxel51 3.0 is free to try at app.voxel51.com.

## What It Is

Voxel51 is a data-centric workflow platform that sits between data collection and model training and evaluation. It turns raw recordings into curated datasets by indexing data for natural language, visual similarity and metadata search. Supported modalities listed include images, video, 3D mesh, lidar, radar, IMU, audio, geolocation and text.

## How the Data Loop Works

- **Search:** query samples, episodes and behaviors with natural language, similarity or metadata.
- **Visualize:** play back images, 3D scenes, sensor streams, maps and annotations in synchronized views.
- **Curate:** slice and filter datasets, using embeddings to weed out low-quality samples.
- **Label:** create and manage labels with automated QA, VLM-assisted labeling and active learning.
- **Evaluate:** surface underperforming classes and failure patterns.

## Voxel51 Agent

The Voxel51 Agent takes natural language requests for data discovery, automated annotation QA, repetitive data work and model evaluation. It can be extended with custom plugins.

## Deployment and Governance

The enterprise edition is described as deployable in cloud, hybrid, on-premise, air-gapped and managed setups. It includes dataset versioning, role-based access controls and extensibility via plugins. The site lists integrations with AWS, Anthropic, Google Cloud, OpenAI, Weights and Biases, Milvus, Hugging Face, NVIDIA and MongoDB. Voxel51 says it was founded in 2018 at the University of Michigan.

## Features
- Natural language, visual similarity and metadata search
- Synchronized multimodal visualization
- Embedding-based data curation
- Automated annotation QA and VLM-assisted labeling
- Active learning
- Model evaluation and failure analysis
- Voxel51 Agent for natural language data workflows
- Custom plugins
- Dataset versioning
- Role-based access controls
- Flexible deployment (cloud, hybrid, on-premise, air-gapped, managed)

## Integrations
AWS, Anthropic, Google Cloud, OpenAI, Weights and Biases, Milvus, Hugging Face, NVIDIA, MongoDB, LeRobot

## Platforms
WEB, API, DEVELOPER_SDK

## Pricing
Freemium — Free tier available with paid upgrades

## Version
3.0

## Links
- Website: https://voxel51.com
- Documentation: https://docs.voxel51.com/
- EveryDev.ai: https://www.everydev.ai/tools/voxel51