BREAD AI
Chat
Host a node

Turn a GPU into a bakery.

Pick your machine. Both paths run the same node, keep 80% of every job (85% on 16 GB+ cards), and are boosted up to 10x when the network is busy. NVIDIA or AMD, 8 GB VRAM or more.

Windows
Gaming PC, desktop app in the tray
  1. 1Download and run the installer (about 45 MB). SmartScreen may warn on first run: More info, then Run anyway.
  2. 2Paste your Solana payout address, or let the app create a wallet for you. It shows the 12-word recovery phrase in Settings.
  3. 3The app detects your GPU, fetches the GPU backend and the model on first run, then goes online.
  4. 4Minimise it to the tray and keep gaming. Jobs only run while you are online.
Download for Windowsx64Version 0.1.0, Devnet preview.

The app makes one outbound connection to BREAD Console and talks to the model runtime on your PC. Prompts in as text, answers out as text. It has no access to your files.

Docker
Linux servers, headless boxes, Docker Desktop
  1. 1
    Install Docker and, for NVIDIA, the NVIDIA Container Toolkit.
  2. 2
    Have a Solana address for payouts, or create one:
    No Solana wallet yet? Create one here. It is generated in your browser and never sent anywhere.
  3. 3
    Run one command:
    docker run -d --name bread-node --gpus all --restart unless-stopped \
  -v bread-node:/data -e BREAD_WALLET=YourSolanaAddress \
  breadaiio/node:latest
  4. 4
    Watch it go online: docker logs -f bread-node. The model downloads into the volume on the first start.
  5. 5
    AMD card: use the breadaiio/node:rocm tag and --device /dev/kfd --device /dev/dri instead of --gpus all.
Optional settings
BREAD_MODELS
bread-chat-8b (default), bread-vision-7b, or both
BREAD_NODE_ID
the name shown in Console; generated when empty
BREAD_REGION
US, CA, SA, EU, UK, AF, ME, IN, ASIA, AU, NZ, Other
BREAD_MAX_JOBS
concurrent jobs, default 1
BREAD_GPU
"NVIDIA RTX 3090,24" if detection fails

Update: docker pull breadaiio/node:latest, remove the container, run the command again. No Docker? The plain bundle and a systemd unit are on the releases page.

Image on Docker Hub

The container reaches the GPU and its own data volume, nothing else on the host. One outbound connection to BREAD Console; prompts in as text, answers out as text.

See the network liveHost FAQ