Helix ML
From a dataset to an experiment
Helix connects a coding agent to a measured ML workflow. Describe a prediction task, upload a labeled CSV, and choose your validation method and run budget. The agent writes candidate models while Helix runs the fits, scores held-out predictions, and records the results. Text, image, and audio assets can accompany the CSV.
A workspace that follows your account
Create a Helix account and connect your own Codex or Claude Code subscription through the native agent's sign-in flow. Railway runs each user's private CPU workspace. The hosted app needs no local terminal, Python environment, or Docker installation. Experiments and agent sign-in state are saved with the workspace.
Inspect the result
Compare trials against a simple baseline, review the generated source, and download the selected model with its configuration and measured results. Helix checks that an exported model reproduces its predictions before accepting it. This is an early CPU preview; GPU acceleration and larger-scale workloads are still future work.