hypha-compute

CLI for deploying and managing compute apps on Hypha Cloud — a Ray alternative with native multi-tenancy.

Installation

npm install -g hypha-compute

Or use directly with npx:

npx hypha-compute --help

Quick Start

# Set your credentials
export HYPHA_TOKEN="your-token"
export HYPHA_WORKSPACE="your-workspace"

# Deploy a Python app
hypha-compute deploy app.py

# Deploy with autoscaling
hypha-compute deploy model.py --min-replicas 1 --max-replicas 10

# List running apps
hypha-compute list

# View logs
hypha-compute logs my-app

# Stop an app
hypha-compute stop my-app

Commands

| Command | Description | |---------|-------------| | deploy <file> | Deploy a .py file or .json manifest | | list | List running compute apps | | status <app-id> | Show detailed app info | | logs <app-id> | View logs for a running app | | stop <app-id> | Stop a running app | | metrics <app-id> | Show deployment metrics | | info | Show connection and worker info |

Deploy

Deploy a Python file directly:

hypha-compute deploy app.py --app-id my-app

The CLI auto-generates a compute-app manifest. If a requirements.txt exists in the same directory, dependencies are included automatically.

Deploy from a JSON manifest for full control:

{
  "type": "compute-app",
  "name": "my-model",
  "entry_point": "main.py",
  "requirements": ["torch", "transformers"],
  "autoscaling": {
    "enabled": true,
    "min_replicas": 1,
    "max_replicas": 5,
    "target_ongoing_requests": 10
  }
}

hypha-compute deploy manifest.json

Options

Global Options

| Flag | Env Var | Description | |------|---------|-------------| | -s, --server-url | HYPHA_SERVER_URL | Hypha server URL (default: https://hypha.aicell.io) | | -w, --workspace | HYPHA_WORKSPACE | Workspace name (default: public) | | -t, --token | HYPHA_TOKEN | Auth token |

Deploy Options

| Flag | Description | |------|-------------| | --app-id <id> | Override app ID | | --name <name> | Override app name | | --min-replicas <n> | Minimum replicas (enables autoscaling) | | --max-replicas <n> | Maximum replicas (enables autoscaling) | | --no-start | Install only, don't start | | --wait-for-service <id> | Wait for service registration (e.g., default) | | --timeout <seconds> | Start timeout (default: 180) |

Python Library

The Python library provides the full compute primitives (autoscaler, P2C router, replica state machine, DAG executor, object store). Install with:

pip install hypha-compute

See the GitHub repository for Python API documentation.

License

MIT