Feya

Install Feya

Run a provider node on your computer to contribute GPU or CPU compute to the network and earn MAGIC. The node speaks to the coordinator, answers inference requests, and keeps itself healthy.

Your pairing code
—

Paste one command, and you're a node.

The installer sets up Python, downloads the Feya app and a ~5 GB model (one-time), and starts your node. No git, no compiler — curl is the only prerequisite. Everything lands under ~/.feya — no system-level changes.

Download Feya for macOS

Disk image for Apple Silicon and Intel Macs. Open it, drag Feya to Applications, launch it from the menu bar, and paste your pairing code. No Terminal needed.

macOS says the app cannot be opened

This build is not notarized by Apple yet, so macOS blocks the first open. Go to System Settings, Privacy & Security, scroll down to Security, and click Open Anyway next to Feya, then confirm. You only do this once; afterwards it opens by double-click like any other app.

curl -fsSL https://feya-ai.com/install.sh | bash -s -- YOUR_PAIRING_CODE
Paste into any shell — curl is the only prerequisite. Have an NVIDIA GPU and Docker? The Docker path is one command too and runs in the background.
Download Feya for Windows

Open FeyaSetup.exe and paste your pairing code when it asks: YOUR_PAIRING_CODE. It sets everything up, no administrator rights needed, and puts the Feya icon in your taskbar tray. Run it again any time to update.

Your browser says it isn't commonly downloaded

Because the installer is new and isn't code-signed yet, Microsoft Edge may hold FeyaSetup.exe in its downloads and say it isn't commonly downloaded. Download it only from feya-ai.com. In Edge's downloads, open the … menu next to it, choose Keep, then Show more and Keep anyway. Other browsers may ask you to choose Keep too.

Windows says it protected your PC

The installer isn't code-signed yet, so Windows SmartScreen stops the first run and shows the publisher as Unknown publisher. Download it only from feya-ai.com. Then click More info, then Run anyway. You only do this once.

or install from PowerShell

$env:PAIRING_CODE="YOUR_PAIRING_CODE"; iwr -useb https://feya-ai.com/install.ps1 | iex
Open PowerShell (Win+S, then "PowerShell"), paste, press Enter. Nothing else to install first; for an NVIDIA GPU, an up-to-date NVIDIA driver is all it needs. The node runs in the background and starts with Windows.
First run takes 5–10 minutes. Most of that is downloading the 5 GB model. The inference engine ships as a pre-built wheel with GPU acceleration baked in, so no local compile is needed. Once your node is up, it registers with the coordinator and appears on your dashboard within a minute. Leave the terminal window open — closing it stops the node.

Stuck?

If the one-liner errors out, try the manual setup instead — it spells out each step so you can see exactly where things go wrong. Help covers the common problems, or write to the Feya team at hello@feya-ai.com with your Support ID (on the Account page) and a screenshot of the error. Don't paste log lines; if we need them, we'll help you make a diagnostics file that removes things like your username, email and internet addresses; you can check it before sending.

This is the from-source path: it clones the repo and compiles the inference engine locally, so it needs git and a C compiler (the quick installer needs neither — it ships a tarball and pre-built wheels). Use this when the one-liner fails and you want to see exactly where, or when you're hacking on Feya itself.

On a Mac, install the Feya app (FeyaAI.dmg, under the quick install above). A node set up from source runs without the app's worker sandbox, so there's no from-source path for macOS here.

1

Prereqs

# Ubuntu / Debian sudo apt-get update sudo apt-get install -y python3.11 python3.11-venv python3-pip git build-essential cmake # Fedora / RHEL sudo dnf install -y python3.11 python3-pip git gcc-c++ cmake
2

Install Feya

git clone https://github.com/audrey-mcknight/pekito.git ~/feya cd ~/feya python3.11 -m venv .venv source .venv/bin/activate # NVIDIA: CMAKE_ARGS="-DGGML_CUDA=on" pip install -e ".[provider]" pip install -e ".[provider]"
3

Download a model

mkdir -p ~/feya/models curl -L -o ~/feya/models/llama-3.1-8b-q4_k_m.gguf \ https://huggingface.co/bartowski/Meta-Llama-3.1-8B-Instruct-GGUF/resolve/main/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf
4

Start the node

PEKITO_COORDINATOR_URL=https://app.feya-ai.com \ PEKITO_NODE_NAME="my-linux-box" \ PEKITO_PAIRING_CODE=YOUR_PAIRING_CODE \ PEKITO_MODEL_PATH=~/feya/models/llama-3.1-8b-q4_k_m.gguf \ pekito-node
1

Prereqs

2

Install Feya

git clone https://github.com/audrey-mcknight/pekito.git $HOME\feya cd $HOME\feya py -3.11 -m venv .venv .venv\Scripts\Activate.ps1 $env:CMAKE_ARGS = "-DGGML_CUDA=on" pip install -e ".[provider]"
3

Download a model

mkdir $HOME\feya\models curl -L -o $HOME\feya\models\llama-3.1-8b-q4_k_m.gguf ` https://huggingface.co/bartowski/Meta-Llama-3.1-8B-Instruct-GGUF/resolve/main/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf
4

Start the node

$env:PEKITO_COORDINATOR_URL = "https://app.feya-ai.com" $env:PEKITO_NODE_NAME = "my-pc" $env:PEKITO_PAIRING_CODE = "YOUR_PAIRING_CODE" $env:PEKITO_MODEL_PATH = "$HOME\feya\models\llama-3.1-8b-q4_k_m.gguf" pekito-node

The Docker path is intended for operators — homelabs, always-on servers, and team deployments where Docker is already part of the environment. If you just want to run Feya on your personal machine, use Quick install.

NVIDIA GPU host

Requires Docker + NVIDIA Container Toolkit. One command: it checks your driver and GPU passthrough, pulls the provider image from ghcr.io, downloads the model, and runs the node in the background with --restart unless-stopped.

curl -sSL https://feya-ai.com/join.sh | FEYA_PAIRING_CODE=YOUR_PAIRING_CODE bash
The script verifies your node actually appears in the provider registry before declaring success. Joining a fleet? Codes are single-use — mint one per box, or set FEYA_IMAGE to your own registry mirror. Prefer to inspect it first? Read join.sh or see the provider quickstart.

Cluster lead + RPC worker profiles

For pipeline parallelism across multiple GPU hosts:

docker compose --profile cluster-lead up --build docker compose --profile rpc-worker up --build
See docker/docker-compose.yml for the full service definitions and env-var mapping.

What happens after install

Your node calls /v1/devices/claim with the pairing code on first start, binds itself to your account, and appears on your dashboard. If the terminal window closes, the node stops — we're shipping a launchd / systemd / Windows service auto-start soon.