Mineshaft is a script you can run locally that allows you to concurrently run miners. You can use this to test out various AI agent architectures easily, see how the incentive mechanism ranks solutions, and more!

​
Features

  • Multi-Miner Support: Run multiple miners with different hotkeys
  • Automatic Port Management: Automatically finds and assigns free ports
  • Interactive Wallet Setup: Guided creation and registration of hotkeys
  • Real-time Logging: Individual log files for each miner and validator
  • Graceful Shutdown: Proper cleanup of all processes on exit

​
Prerequisites

  • .venv set up to run miners and validators
    • Make sure SWE-agent is installed, for running miners
  • A registered keypair for the validator
  • A miner coldkey with enough TAO to register multiple hotkeys

​
Installation

pip install -r requirements-mineshaft.txt

​
Usage

  1. Ensure the validator runs (you have a validator coldkey+hotkey)
  2. Ensure miners work (installed pip packages, SWE-agent works, have a miner coldkey)
  3. Run Mineshaft:
python mineshaft.py

Mineshaft will:

  • Check for required configuration
  • Offer to create, register, and post miner hotkeys if needed
  • Start the specified number of miners
  • Start the validator
  • Create individual log files for each process

​
Logs

Logs are stored in the logs_mineshaft directory, with a timestamped folder for each run:

logs_mineshaft/
└── YYYYMMDD_HHMMSS/
    ├── miner_1.log
    ├── miner_2.log
    ├── miner_3.log
    └── validator.log

​
Process Management

  • All processes are properly managed and cleaned up on exit
  • Use Ctrl+C to gracefully shut down all processes
  • Each miner runs on its own port, automatically assigned
  • The validator runs with the configured wallet settings

​
Notes

  • The coldkey wallet must exist before running Mineshaft
  • Hotkeys will be created and registered interactively if they don’t exist
  • Each miner gets a unique port, starting from BASE_MINER_PORT and counting down
  • The external IP is automatically detected on macOS, falls back to 0.0.0.0 otherwise