tup-cli 0.1.0

Run any command in the cloud. Close your laptop, keep running.

tup

Run any command in the cloud. Close your laptop, keep running.

# Local
uv run python train.py

# Cloud (just prefix with tup)
tup uv run python train.py

Installation

pip install tup

Quick Start

# Configure (one-time)
tup config set api_url https://tup-api.your-account.workers.dev

# Run a command in the cloud
tup python train.py --epochs 100

# Detach and check later
tup --detach python train.py --epochs 1000
tup logs        # Stream logs
tup status      # Check job status

Usage

tup <command>                    # Run command, stream logs until done
tup --detach <command>           # Run in background, return job ID
tup --name "exp-v1" <command>    # Named job for easy reference

tup logs [job-id]                # Stream logs (default: latest job)
tup status                       # List all jobs
tup stop <job-id>                # Kill a running job
tup attach <job-id>              # Reattach to running job's logs

Options

--detach, -d          Run in background
--name, -n NAME       Job name
--instance TYPE       Instance size: lite, basic, standard-1, standard-2, standard-3, standard-4
--env KEY=VALUE       Pass environment variable (can repeat)
--timeout DURATION    Max runtime (e.g., "2h", "30m", "1d")

Configuration

Configuration is stored in ~/.tup/config.toml:

[default]
api_url = "https://tup-api.your-account.workers.dev"

[env]
# Environment variables passed to all jobs
WANDB_API_KEY = "your-key"
TINKER_API_KEY = "your-key"

Or use environment variables:

export TUP_API_URL="https://tup-api.your-account.workers.dev"

How It Works

1. tup packages your current directory (respecting .gitignore)
2. Uploads the bundle to Cloudflare R2
3. Spins up a container with your command
4. Streams logs back to your terminal
5. Container runs until your command exits

Self-Hosting

Deploy your own tup backend:

cd worker
cp wrangler.example.toml wrangler.toml
# Edit wrangler.toml with your account details
pnpm install
pnpm run deploy

See CONTRIBUTING.md for architecture details.

Instance Types

Type	vCPU	Memory	Disk	~Cost/hour
lite	1/16	256MB	2GB	$0.002
basic	1/4	1GB	4GB	$0.01
standard-1	1/2	4GB	8GB	$0.04
standard-2	1	6GB	12GB	$0.08
standard-3	2	8GB	16GB	$0.16
standard-4	4	12GB	20GB	$0.32

License

MIT