Platform API

You provide four things; Kite does the rest. There is no infrastructure to manage — generation runs on our GPUs, episodes are produced in parallel, and the finished dataset is delivered in standard LeRobot Parquet format, ready to train on.

The Platform API lives at https://api.kiteml.com/v1. Authenticate every request with an API key (see API Keys) as a Bearer token. Create a key scoped to augmentations:

# Create a scoped key from the dashboard (Platform API > API keys),
# or via the API, then export it:
export KITE_API_KEY=kite_aBcDeFgH...

# Verify it works:
curl https://api.kiteml.com/v1/keys/me \
  -H "Authorization: Bearer $KITE_API_KEY"

One call starts a run. Give it the source dataset, your prompt, the episode count, and where the results should go (download keeps them on Kite for you to fetch; huggingface pushes to your account).

curl -X POST https://api.kiteml.com/v1/augmentations \
  -H "Authorization: Bearer $KITE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "source": { "repo_id": "lerobot/pusht" },
    "instructions": "change the table surface to white marble, vary the lighting",
    "config": { "episode_count": 100 },
    "output": { "type": "download" }
  }'

You get back the augmentation resource, including its id, immediately:

{
  "id": "aug_01J8X4M2K9ZQ6R7T3V5W8Y0B1C",
  "object": "augmentation",
  "status": "queued",
  "progress": 0.0,
  "output": { "type": "download" },
  "created_at": "2026-07-21T09:14:00Z"
}

Episodes are generated and saved incrementally. Poll the run to watch it move through queued → processing → succeeded, with a live progress value and a human-readable status message:

curl https://api.kiteml.com/v1/augmentations/aug_01J8X4... \
  -H "Authorization: Bearer $KITE_API_KEY"

# { "status": "processing", "progress": 0.42,
#   "status_message": "Generated 42 of 100 videos" }

For long runs, register a webhook instead of polling — Kite sends a signed augmentation.completed (or .failed) event to your URL when the run finishes. Every event is also readable at GET /v1/events, so polling is always a fallback.

When status is succeeded, a download run exposes its files. Fetch each one, preserving its path, to reconstruct a standard LeRobot Parquet dataset on disk:

"output": {
  "type": "download",
  "files": [
    { "path": "meta/info.json",                 "bytes": 3186,  "url": "https://..." },
    { "path": "data/chunk-000/file-000.parquet", "bytes": 16457, "url": "https://..." },
    { "path": "videos/chunk-000/observation.images.cam/episode_000000.mp4", "url": "https://..." }
  ]
}

The kite augment download CLI command does this for you — see the CLI docs. If you chose huggingface output instead, output.url links the dataset pushed to your account.

Full request/response schemas are published as OpenAPI at https://api.kiteml.com/v1/openapi.json (rendered at /v1/docs).

The same API trains policies: POST /v1/training_runs fine-tunes one or more LeRobot policies on managed GPUs, billed from the same credit balance. See Training runs for the full lifecycle.