Use Voxel51 3.0's free tier and Agent to streamline robotics data annotation and evaluation workflows
The new platform version combines search, annotation, and evaluation tools into one system, adding an AI agent and free access tier for robotics developers.
Key points
- Voxel51 3.0 merges data curation, annotation, and evaluation into a single platform for physical AI teams.
- The release introduces Voxel51 Agent with over 45 MCP tools and a free self-service tier with monthly usage limits.
- New features include time-series multimodal support, unified annotation across data types, and agentic labeling capabilities.
Voxel51 has released version 3.0 of its data platform, combining search, curation, annotation, and evaluation into one system for teams working with physical AI. According to MarkTechPost, this update targets groups training models on sensor data from robots, vehicles, and factories. The release adds a new AI agent, a data provenance layer, and free self-service access for users.
In plain English
The platform now supports time-series multimodal data, allowing teams to view multiple sensor streams on one timeline. It also features unified annotation that covers 2D, 3D, video, and multimodal data with automated quality checks. Agentic labeling uses vision-language models to generate labels from natural-language instructions and examples.
The background
Voxel51 Search uses a persistent hybrid index that automatically computes embeddings so search works immediately. Semantic retrieval finds similar moments without shared labels, while lexical queries handle exact terms and metadata. Vector retrieval runs on LanceDB, supporting vector, scalar, and full-text indexes. Teams can choose what to embed and when, with incremental updates avoiding full rebuilds.
What changes now
Voxel51 Agent chains search, curation, annotation, and evaluation into repeatable workflows. A robotics team can search behaviors across thousands of episodes and curate balanced training sets. Perception teams can start from model failures and pull similar long-tail scenarios for labeling. The free tier includes 30,000 inference requests and 30 compute hours monthly, but caps storage at 20 GB.
Key terms explained
Multimodal data combines different sensor types like video, audio, and 3D scans into one dataset. Teams usually align these streams to analyze complex environments. Data provenance tracks the origin and history of each data point, ensuring transparency. Agentic workflows use AI to automate repetitive tasks like labeling or searching. Check whether your current tools support these integrated features to reduce manual effort and improve dataset quality for physical AI projects.
What to do and how to stay safe: Voxel51
- Review your current data curation tools to see if they support unified multimodal annotation and search.
- Evaluate whether the free self-service tier meets your team's storage and compute needs for initial testing.
- Check if your existing workflows can integrate with MCP tools to automate repetitive data tasks.
- Monitor for updates on dataset versioning and data provenance features to ensure compliance and traceability.
Step-by-step guide: Stop AI Chatbots Leaking Sensitive Business Data
General security guidance from the Vector Update newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What new features does Voxel51 3.0 offer for physical AI teams?
It adds Voxel51 Agent with 45+ MCP tools, time-series multimodal support, unified annotation, and agentic labeling.
Is there a free tier available for Voxel51 3.0?
Yes, the free tier includes Voxel51 Agent, 30,000 inference requests, and 30 compute hours per month, with storage limits.
Can Voxel51 3.0 be deployed on-premise?
Yes, it runs on a hosted web app or can be self-hosted in cloud, hybrid, on-prem, and air-gapped setups.
