Voxel51 has released Voxel51 3.0, a very important update to its data platform. Voxel51 3.0 is an end-to-end multimodal data platform that puts search, curation, annotation and evaluation in one system. It targets teams training models on sensor data from robots, vehicles and factories. The release adds Voxel51 Agent, a data provenance layer and free self-service access.
TL;DR
- Scale: Voxel51 Agent ships with 45+ MCP tools, 80+ operators and 18 open-source skills.
- Runs on: Hosted web app at app.voxel51.com, or self-hosted in cloud, hybrid, on-prem and air-gapped setups. No local GPU requirement disclosed.
- Performance: No product benchmarks published. Voxel51 cites its own survey data instead.
- Best: Free tier includes Voxel51 Agent, 30,000 inference requests and 30 compute hours per month.
- Worst: Free tier caps at 3 datasets of 10,000 samples each and 20 GB of storage.
What is new in Voxel51 3.0?
Voxel51 3.0 expands the platform beyond its roots in visual data with 3 core additions.
- Time-series multimodal support: Teams can view multiple sensor streams on 1 timeline. Segment-level embeddings find specific moments inside long recordings.
- Unified annotation: Labeling covers 2D, 3D, video and multimodal data. It includes automated QA, annotator metrics and human-in-the-loop review.
- Agentic Labeling: Vision-language models generate labels from natural-language instructions and examples.
Supported modalities include MCAP, images, video, 3D point clouds, meshes, DICOM, NIfTI, geospatial data and audio.
How does Voxel51 Search work?
Voxel51 Search is a persistent hybrid index. It indexes datasets and computes embeddings automatically, so search works from day 1.
Semantic retrieval finds visually or semantically similar moments, even without shared labels. Lexical and structured queries handle exact terms, timestamps, sensor fields and metadata. Embeddings and structured fields stay in native form and are combined at retrieval time.
Vector retrieval runs on LanceDB. It supports vector, scalar and full-text indexes, and separates storage from compute. Teams choose what to embed, which model and compute to use, and when. Incremental updates avoid full rebuilds when samples arrive or schemas change.
What can Voxel51 Agent do?
Voxel51 Agent chains search, curation, annotation and evaluation into repeatable workflows.
Example workflows:
- A robotics team searches behaviors across thousands of episodes, then curates balanced training sets.
- A perception team starts from a model failure and pulls similar long-tail scenarios for labeling.
The agent can also write code to build custom plugins. All workflows inherit the organization’s permissions and governance controls.
How does data provenance work in Voxel51 3.0?
Voxel51 3.0 adds 4 provenance layers:
- Deployment-level activity log: who did what, and when, across the deployment.
- Dataset-level activity log: how a single dataset evolved.
- Sample-level edit history: label, metadata and property changes per sample.
- Dataset versioning: immutable snapshots to reproduce training and evaluation runs.
Events flow through fiftyone-mq for job scheduling and Activity Core for event tracking.
Can teams bring their own models?
Yes. The platform is model-agnostic across embeddings, auto-labeling, inference and evaluation. A team can bootstrap with a foundation model, then swap in its own domain model.
Why does Voxel51 argue data work decides outcomes?
Voxel51 cites its 2026 State of Physical AI survey. 58% of “exceptional” teams spend more than half their project time on data work. Only 21% of struggling teams do the same. This is vendor-run research, so read it as directional.