TAH User Guide
Plug in to power decentralised AI model training
Training at Home (TAH) application allows you to connect and partake in our decentralized AI training platform IOTA. You can plug in any hardware you have access to and earn rewards for powering decentralized AI model training.
Installation
To become a first beta tester of TAH application fill the form at iota.macrocosmos.ai/train-at-home. We are scaling the system and gradually rolling out to more users. Currently TAH only supports MacOS, with Linux support coming soon. Look at Hardware & OS Requirements for more details.
If you already have access, you can double click a .dmg file to install the app.
Press Open in the push notification:
Video below demonstrates installation process
Operating TAH
Once the app is installed you should be able to see the main dashboard:
You’ll see metrics for your model, and in the center there will be a visualization of all the other participants in the training.
Main controls (top right of the screen)
Start training: begins contributing your GPU/CPU to the current run. Click again to stop/exit the run.
Connect: optionally paste your wallet coldkey to receive rewards to that address. You can train without connecting; if you connect later, rewards accrue to the provided coldkey from that point forward.
Status pill: shows whether your hardware is ready (e.g., “Connected • Your GPU is ready to contribute”). Resolve any warnings before starting.
Start a session
Launch the app and wait for the status pill to read Connected.
(Optional) Click Connect, paste your wallet coldkey, and save.
Click Start training. The run ID, model size, and layer count populate in the right panel; progress and tokens/hour graphs begin updating.
To stop, click Start training again (it toggles off) or quit the app.
Monitor while training
Run panel: shows run ID, model size, layers, cumulative progress, and tokens/hour graph.
Network view: visualizes peers and traffic; switch tabs (Network/Layer/Miner) for different views.
Loss chart: training loss over time; toggle Log/Linear and tokens/time axes.
Run Log (bottom right): per-epoch/step logs; lock icon indicates read-only when idle.
Resource behavior
The app uses available GPU/CPU once training starts. If you need to pause, toggle Start training off.
Keep the laptop powered and on a stable network for steady rewards. Thermal throttling may reduce contribution; use a cooler or lower-power profile if needed.
Updates
If an update is available, install the latest build before starting a new session for compatibility with the active run.
Rewards
Beta Launch is operating with o rewards. Once the major version of TAH is published, the earnings will be calculated and paid as described below.
How Rewards Are Calculated
Rewards are primarily based on:
The amount of tokens you contribute during training.
How many times you contribute your model weights back to the network.
Additional factors such as uptime, quality, and current network parameters.
Note: Exact weighting can evolve as the system updates.
Payout Cadence
Payouts occur approximately every 24 hours; this window may be extended in the future.
Minimum payout is 2 ALPHA. Balances below the threshold roll over to the next payout.
Network transfer costs are deducted from the amount you receive.
To receive payouts, optionally connect a wallet by pasting your public coldkey (via the Connect button in the app). If you don’t connect, training still runs but rewards won’t be delivered to you; unconnected earnings are not paid retroactively.
Viewing Earnings
In the app, click the Miner button (top left) to view your current earnings and contribution stats.
Taxes & Compliance
You are responsible for complying with local tax and regulatory requirements related to rewards.
For more questions use FAQs
If you have any issues use TAH Support
Last updated