Open and index
A streaming SQLite sidecar so a 50,000-episode LeRobot dataset browses without loading media into memory.
→The local-first IDE for reviewed teleop and VLA datasets. Open LeRobot, RLDS, OpenX, HDF5, ROS bag, and mp4 captures. Scrub, tag, cluster, probe, and export.
LeRobot v3/v2, RLDS, OpenX, HDF5, ROS bag, mp4 + jsonl — one adapter contract.
Video, depth, joint state, force/torque, audio, language — in one timeline.
Hugging Face, local disk, embodiment card, sensor QA, failure clusters.
Open the dataset. Review the runs. Export the subset training can actually use.
A streaming SQLite sidecar so a 50,000-episode LeRobot dataset browses without loading media into memory.
→Apply failure taxonomy tags, intervention labels, sensor anomaly notes, and cluster-wide training decisions.
→Run VLA perturbation probes. Generate embodiment cards. Export to Hugging Face, disk, or AuraOne intake.
Training-ready files. Local-first. The next reviewer picks it up without a hosted account.
Episodes filtered by tag, cluster, or training-readiness score. Hugging Face or local disk.
Markdown card for robot morphology, sensors, action spaces, coordinate frames, teleop method, and limits.
Timestamp drift, video reference health, action and state field checks, intervention markers.
CLIP, hash, or custom embedding clusters with representative thumbnails and cluster-wide tags.
Signed .auraonepkg with a privacy preview before AuraOne Robotics Programs handoff.
Local-first IDE for MCP and A2A agents. Replay, compare, export.
See the page →Git-native criterion-level authoring with calibration and diff.
See the page →lerobot-quality-gates, robot-recovery-bench, vla-robustness-kit, embodiment-card.
See the page →The IDE for reviewed datasets, on your machine. No private data leaves disk unless you say so.