Voxel51 3.0 unifies multimodal data tools for physical AI teams

Voxel51, a company focused on computer vision data tooling, has launched version 3.0 of its platform. Announced on October 18, 2026, the release is aimed at teams building physical AI, such as robotics and autonomous systems. According to the company, it consolidates the core workflows for multimodal sensor data into a single environment.

In Voxel51 3.0, searching, curating, annotating and evaluating data are combined in one product. The goal is to let teams work with data from cameras, lidar and other sensors without shuttling it between separate tools.

An agent, versioning and a free tier

A headline addition is Voxel51 Agent, an AI assistant that the company says ships with more than 45 MCP tools. MCP, the Model Context Protocol, is an open standard that lets AI models connect to external tools and data sources in a consistent way.

The platform also gains dataset versioning, which helps teams track changes to training and test data and roll back to earlier states when needed. Voxel51 is additionally offering a free self-service tier that users can sign up for on their own.

Why it matters

AI systems that operate in the physical world depend on large, messy datasets gathered from many sensors. Finding, cleaning and labeling that data, then testing models against it, often requires a patchwork of tools and consumes much of a team's effort.

Voxel51 is known in the computer vision community for its open-source FiftyOne tool. With this release, the company is pushing further toward an end-to-end data platform rather than standalone utilities. Pricing details beyond the free tier and independent assessments of how well the agent performs have not yet been made public.