sami-cli 0.1.0

Python SDK and CLI for SAMI Dataset Distribution Platform

SAMI CLI

Python SDK and CLI for the SAMI Dataset Distribution Platform. Upload, download, and manage robotics datasets in LeRobot format.

Note: Dataset uploads require platform admin privileges (globalRole: platform_admin). Regular users can browse and download datasets but cannot upload.

Installation

pip install sami-cli

For development:

cd sami-cli
pip install -e ".[dev]"

Quick Start (CLI)

# Login (credentials saved to ~/.sami/)
sami login
Email: [e-mail]
Password: ********
Logged in as [e-mail]
  Organization: Acme Robotics

# List datasets
sami list

# Upload a dataset (requires admin)
sami upload ./my_dataset --name "Robot Arm Demo"

# Download a dataset
sami download <dataset-id> --output ./downloaded

# Show your info
sami whoami

# Logout
sami logout

CLI Reference

Command	Description
sami login	Authenticate and save credentials
sami logout	Clear saved credentials
sami whoami	Show current user info
sami config	View/set configuration
sami list	List accessible datasets
sami upload <path>	Upload a LeRobot dataset
sami download <id>	Download a dataset
sami info <id>	Show dataset details
sami delete <id>	Delete a dataset

Command Options

# Upload with options
sami upload ./dataset \
    --name "My Dataset" \
    --description "Kitchen manipulation tasks" \
    --task-category manipulation \
    --workers 8

# Download with options
sami download abc123 \
    --output ./my_data \
    --workers 8

# List with filters
sami list --status ready --limit 50

# Set custom API URL
sami config --api-url https://api.example.com/api/v1

Environment Variables

For CI/CD pipelines, you can use environment variables instead of sami login:

Variable	Description
SAMI_API_URL	Override API URL
SAMI_ACCESS_TOKEN	Use token directly (skip login)
SAMI_EMAIL	Email for login
SAMI_PASSWORD	Password for login

# Example: CI/CD usage
export SAMI_ACCESS_TOKEN="your-jwt-token"
sami list
sami download abc123

Python SDK

Using Saved Credentials

After running sami login, use credentials in Python:

from sami_datasets import SamiClient

# Use saved credentials from ~/.sami/
client = SamiClient.from_saved_credentials()

# List datasets
datasets = client.list_datasets()
for ds in datasets:
    print(f"{ds.name}: {ds.episode_count} episodes")

Direct Authentication

from sami_datasets import SamiClient

# Authenticate directly
client = SamiClient(
    email="[e-mail]",
    password="your-password",
)

# Upload a LeRobot dataset
dataset = client.upload_dataset(
    name="my-dataset-v1",
    path="/path/to/lerobot/dataset",
    description="Kitchen manipulation tasks",
    task_category="manipulation",
)
print(f"Uploaded: {dataset.id}")

# Download a dataset
client.download_dataset(
    dataset_id=dataset.id,
    output_path="./downloaded_dataset",
)

API Methods

# Authentication
client.login(email, password)
client.get_current_user()

# Datasets
client.list_datasets(page=1, limit=20, status=None)
client.get_dataset(dataset_id)
client.upload_dataset(name, path, description=None, task_category=None, max_workers=4)
client.download_dataset(dataset_id, output_path, max_workers=4)
client.delete_dataset(dataset_id)

# Sharing
client.assign_dataset(dataset_id, organization_id, permission_level)
client.remove_assignment(dataset_id, assignment_id)

LeRobot Format

Datasets must be in LeRobot format:

my_dataset/
  meta/
    info.json       # Required: episodes, frames, fps, features
    stats.json      # Optional: statistics
    episodes/       # Optional: episode metadata
  data/
    chunk-000/      # Parquet files with episode data
    chunk-001/
  videos/           # Optional: video files
    chunk-000/

The meta/info.json must contain:

• total_episodes: Number of episodes
• total_frames: Total frame count
• fps: Frames per second

LeRobot Integration

Downloaded datasets work directly with LeRobot:

from lerobot.common.datasets.lerobot_dataset import LeRobotDataset

# Load downloaded dataset
dataset = LeRobotDataset("./my_dataset")

# Use in training
for batch in dataset:
    observation = batch["observation.state"]
    action = batch["action"]
    # ... train your model

Dataset Object

@dataclass
class Dataset:
    id: str
    name: str
    description: Optional[str]
    task_category: Optional[str]
    robot_type: Optional[str]
    episode_count: Optional[int]
    total_frames: Optional[int]
    fps: Optional[float]
    file_size_bytes: int
    upload_status: str  # pending, uploading, processing, ready, failed
    created_at: datetime
    organization_name: str
    features: Optional[Dict[str, Any]]
    assignments: List[Dict[str, Any]]

Exceptions

from sami_datasets import (
    SamiError,              # Base exception
    AuthenticationError,    # Login failed
    NotFoundError,          # Resource not found
    PermissionDeniedError,  # Access denied
    UploadError,            # Upload failed
    DownloadError,          # Download failed
    ValidationError,        # Invalid dataset format
)

Requirements

• Python >= 3.9
• requests >= 2.28.0
• tqdm >= 4.65.0