🖥️ Run an AI Employee on Your Own Hardware: Setup Guide

Connect a Linux computer with an NVIDIA GPU to VEP, and an AI employee answers with an AI model installed on it. Its memory, files and chat history stay in VEP Cloud. This is in early access: VEP switches it on for selected companies.

In the app, your computer is called a node or a resource.

Use the text pairing code, not just the QR. Wherever VEP shows a pairing QR for your computer, it also shows a text code. For terminal setup, always run vep-node pair YOUR_CODE with the text code shown in Compute Resources.

Prerequisites

Requirement	Minimum	Recommended
OS	Ubuntu 22.04+	Ubuntu 24.04 LTS
GPU	NVIDIA GTX 1080 (8GB)	RTX 4090 (24GB) or better
RAM	16 GB	32 GB+
Storage	50 GB free	200 GB+ SSD
Network	10 Mbps	100 Mbps+
Node.js	22+	22 LTS

Quick Start

1 Install the VEP Node Agent

mkdir -p "$HOME/vep/private-vault"
curl -fsSL https://vep.live/install-node.sh | bash -s -- --vault-root "$HOME/vep/private-vault"

This installs the agent into $HOME/.vep-node-agent, creates the CLI wrapper $HOME/.local/bin/vep-node, initializes the agent's configuration, and runs a readiness check.

If this computer is already paired and you want one command that updates the agent, installs a persistent service, and brings the agent back online after logout/reboot, use a root-driven install for the target Linux user:

sudo bash -lc 'curl -fsSL https://vep.live/install-node.sh | bash -s -- --service-user vepuser --vault-root "/home/vepuser/vep/private-vault" --install-service'

If your shell does not already include $HOME/.local/bin in PATH, run:

export PATH="$HOME/.local/bin:$PATH"

Or manual install:

mkdir -p /tmp/vep-node-agent
curl -fsSL https://vep.live/downloads/vep-node-agent.tar.gz | tar -xz -C /tmp/vep-node-agent --strip-components=1
cd /tmp/vep-node-agent
bash setup/install-node-agent.sh
vep-node init --server https://vep.live --vault-root "$HOME/vep/private-vault"

2 Register Your Computer in VEP

Open vep.live → Add panel → Compute Resources → + Add Resource

In the Add Compute Node window, enter a name (e.g. "Home GPU") and click Create Node. The pending resource card shows a pairing code and the exact command to run. The code works for 10 minutes.

On a phone: Dashboard → Usage & Billing → Integrations → On-prem / BYOC → Open BYOC pairing.

3 Pair Your Computer

vep-node pair ABC123

Replace ABC123 with the actual text pairing code from the VEP dashboard.

Pairing is one-time per local config. If you move to a different Linux user or a different computer, create a new pairing code or migrate the full ~/.vep-node state.

4 Start the Agent

vep-node start

Your computer should appear as Online in the VEP dashboard within seconds.

When the agent starts you should see a line like VEP Node Agent v0.x.y. That version line is the fastest way to confirm which agent release is actually running.

5 Install as Service (Optional)

sudo "$HOME/.local/bin/vep-node" install-service --user "$USER"

This creates a systemd service that auto-starts on boot. If the computer will run unattended, this is strongly recommended.

Equivalent unattended one-liner after pairing:

sudo bash -lc 'curl -fsSL https://vep.live/install-node.sh | bash -s -- --service-user vepuser --vault-root "/home/vepuser/vep/private-vault" --install-service'

GPU Setup

NVIDIA Drivers

sudo apt install nvidia-driver-550 nvidia-utils-550
nvidia-smi  # verify GPU detected

Ollama (for local LLM inference)

curl -fsSL https://ollama.com/install.sh | sh
# GPU memory 12 to 23 GB
ollama pull qwen3:14b
# GPU memory 8 to 11 GB
# ollama pull qwen3:8b
# GPU memory 24 GB or more
# ollama pull qwen3:32b

Pull only the model for your GPU memory: to use a commented line, run it without the #. In VEP you then choose that model under Runtime Controls and click Assign Model (see Enable Local Compute below).

To verify Ollama is ready:

curl -fsS http://localhost:11434/api/tags

Enable Local Compute

To put a new AI employee on the resource, create it first: in Fleet Dashboard or Employee Cards, click + New Employee, enter a name, choose a role and click Create. The new employee is active right away and appears in the Assign Brain list.

In VEP Dashboard → Compute Resources:

open your resource
under Assign Brain, choose the employee and click Assign Brain
open the resource again and click Apply Recommended Local Setup, or pick a runtime and click Set Employee Runtime
under Runtime Controls, choose one of the local AIs installed on the computer and click Assign Model
the employee then runs on the resource; its knowledge, files and private data stay in VEP Cloud, and while the resource is offline the employee does not answer. In chats with your customers, VEP Cloud also uses cloud AI models for some replies and steps, for example to answer a greeting, even while the resource is offline

When routing is active, Boss Chat for that employee should show the resource name and 🖥️ Your Hardware instead of ☁️ VEP Cloud.

Routing is opt-in per employee: only employees placed on a resource (a preferred resource, or Preferred Execution set to Run on this node) run there; every other employee of the company keeps running in VEP Cloud even when "Prefer Local Compute" is on. Assign Brain alone does not move the employee: it keeps answering from VEP Cloud until you click Apply Recommended Local Setup or Set Employee Runtime. After either button the employee runs only on the resource and never falls back to VEP Cloud: if it does not answer, check that the resource is online. If the resource is online and the employee still does not answer, local compute may not be enabled for your company yet: ask VEP Support.

💡 Tip: Local inference costs €0 per token: VEP does not charge for answers from your computer. You pay for the computer and its power, and replies to your customers still count toward your plan's monthly message limit.

Troubleshooting

The computer won't connect

Check your firewall allows outbound connections to vep.live:443
Run vep-node status to see current connection state
Check logs: journalctl -u vep-node -f

`vep-node` command not found

Run export PATH="$HOME/.local/bin:$PATH"
Then retry vep-node status

GPU not detected

Run nvidia-smi — if it fails, reinstall NVIDIA drivers
Reboot after driver installation

Pairing code expired or you need the text code

Codes expire after 10 minutes
The pending resource card shows the text code and the exact vep-node pair command to run; no QR is needed
Click Generate New Code on the pending resource card to get a new one

Auto-update has not applied yet

The agent checks for updates automatically
The running version is shown in the first line of vep-node start
If the VEP dashboard says Waiting for Latest Node Agent, the agent on your computer is still an older version, and newer runtimes such as OpenHands cannot be enabled yet

⚠️ Security Note: The agent's private key is generated on your computer and never leaves it; VEP receives only its public key. VEP Cloud still builds every request it sends to your computer and keeps every answer.

VEP — Virtual Employee Platform · vep.live

🖥️ Run an AI Employee on Your Own Hardware: Setup Guide

Prerequisites

Quick Start

GPU Setup

NVIDIA Drivers

Ollama (for local LLM inference)

Enable Local Compute

Troubleshooting

The computer won't connect

vep-node command not found

GPU not detected

Pairing code expired or you need the text code

Auto-update has not applied yet

`vep-node` command not found