specter-host

Remote GPUs, locally. Run python train.py on cloud GPUs as if they were local.

Install

# Recommended (isolated install)
pipx install specter-host

# Or with pip
pip install specter-host

Quick Start

# Create and activate a virtual environment
python -m venv .venv && source .venv/bin/activate

# Authenticate via browser
specter login

# Provision a GPU — patches your active venv
specter install h100

# Run Python as usual — it transparently executes on the remote GPU
python train.py

# pip install works locally AND syncs to the remote automatically
pip install transformers

# When done, tear down the pod and restore your venv
specter uninstall

How It Works

1. specter install h100 provisions a cloud GPU pod via the Specter broker
2. Your active venv is patched — bin/python becomes a transparent proxy
   • The original binary is backed up as python.specter-original
   • pip, pytest, and other entry points continue to run locally
3. When you run python train.py, Specter:
   • Rsyncs your workspace to the remote machine
   • Executes the command via SSH
   • Streams stdout/stderr back in real-time
   • Returns the remote exit code
4. Dependencies stay in sync automatically:
   • At install time: pip freeze → remote pip install
   • After every pip install: auto-synced to the remote pod
5. specter uninstall restores your venv to its original state

License

MIT