Scattered GPUs
Gaming PC in bedroom, workstation in office, old server in closet. All idle most of the time.
Free Forever • No Credit Card • No Limits
Turn All Your Machines into
One Local AI
Gaming PC + Mac + old server. rbee orchestrates your machines as one hive. 5-minute setup. Use all your GPUs across all your machines. Free forever.
- Multi-machine orchestration
- Heterogeneous hardware support
- Free forever
RTX 4090 + Mac M2 Ultra + RTX 3090 = one unified colony
Multi-Machine Setup
Connect all your hardware
Free Forever
Get the Free Setup Guide
5-minute install, SSH-based deployment, works with all your hardware. Gaming PC + Mac + Server = one unified hive.
Manual Setup Hell
Different OS versions, different CUDA drivers, different Python environments. Nothing works together.
No Orchestration
Want to run a 70B model? Good luck manually splitting it across machines and keeping track of processes.
Wasted Hardware
Thousands of dollars in GPUs sitting idle because there's no easy way to use them together.
Unified Orchestration
One command to deploy models across all your machines. rbee handles SSH, GPU detection, and load balancing.
SSH-First Lifecycle
No agents, no daemons. Pure SSH control. Start, stop, monitor—all from your terminal.
Use All Your Hardware
CUDA, Metal, CPU—rbee detects and uses whatever you have. Mix and match architectures freely.
Complete Privacy
Zero telemetry. Zero tracking. Your data never leaves your network. Open-source and auditable.
1
Install rbee-keeper
Install on your main machine
# Ubuntu/Debiancurl -fsSL https://rbee.sh/install.sh | bash
# macOSbrew install rbee/tap/rbee-keeper
# Verify installationrbee-keeper --version2
Add Your Machines
Register remote machines via SSH
# Add a machine to your poolrbee-keeper pool add \ --name gaming-pc \ --host 192.168.1.100 \ --ssh-key ~/.ssh/id_ed25519
# Add multiple machinesrbee-keeper pool add --name workstation --host 192.168.1.101rbee-keeper pool add --name mac-mini --host 192.168.1.102
# List all machinesrbee-keeper pool list3
Deploy a Model
Run models across your homelab
# Deploy a model across available GPUsrbee-keeper infer \ --model llama-3.1-70b \ --pool gaming-pc,workstation \ --auto-balance
# rbee automatically:# - Downloads the model if needed# - Detects available GPUs# - Splits model across machines# - Starts inference server4
Monitor & Manage
Monitor your homelab
# Check statusrbee-keeper status
# View GPU utilizationrbee-keeper gpu-info --all
# Stop a modelrbee-keeper stop --model llama-3.1-70b
# Clean shutdown (no orphaned processes)rbee-keeper shutdown --pool gaming-pcPool Registry Management
Configure remote machines once. rbee-keeper handles SSH, validates connectivity, and keeps your pool registry synced.
$ rbee-keeper pool list
● gaming-pc (192.168.1.100) - RTX 4090 24GB - READY
● workstation (192.168.1.101) - RTX 3090 24GB - READY
● mac-mini (192.168.1.102) - M2 Max 32GB - READY
$ rbee-keeper infer --model llama-3.1-70b --auto-balance
→ Detected 3 machines with 72GB total VRAM
→ Splitting model across gaming-pc + workstation
✓ Model loaded in 12.3s
✓ Inference server ready at http://localhost:8080
CheckCircle
Auto-Discovery
Detects GPUs, VRAM, and capabilities automatically
Zap
Smart Balancing
Optimizes model placement across available hardware
Shield
SSH Security
Uses your existing SSH keys and known_hosts
Automatic Provisioning
rbee-keeper orchestrates model deployment across your homelab via SSH
rbee-keeper
rbee-keeper
(orchestrator)
SSH: Deploy model shard 1/2
gaming-pc
gaming-pc
RTX 4090
SSH: Deploy model shard 2/2
workstation
workstation
RTX 3090
All communication via SSH
No agents or daemons required
Multi-Backend GPU Support
CUDA, Metal, and CPU backends. rbee detects your hardware and lets you choose explicitly.
Prohibited:
No silent CPU fallbackNo hidden backend switches
What happens:
Explicit backend selectionClear hardware detectionHelpful error messages
GPU not detected or insufficient VRAM
- Check GPU drivers are installed (nvidia-smi for CUDA)
- Use --backend cpu to run on CPU explicitly
- Try a smaller quantized model (Q4_K_M)
rbee detect — homelab-server.local
✓ CUDA detected: RTX 3080 (10GB VRAM)
✓ CPU detected: 16 cores available
✗ Metal: not available (Linux)
Available backends:
cuda × 1cpu × 1metal × 0
Total devices: 2
Hardware detection cached for fast startup.
Cpu
NVIDIA CUDA
RTX 20/30/40 series, Tesla, A100, H100. Automatic detection and VRAM checks.
Laptop
Apple Metal
M1, M2, M3 chips. Native Metal acceleration on Apple Silicon.
Server
CPU Fallback
No GPU? Run on CPU with optimized quantization (Q4, Q8).
Power Cost Calculator
20h
1h24h
≈ 600h / mo
80%
10%100%
Monthly Power Cost
Monthly
€216.00
Based on 20h/month at 80% load, €0.30/kWh
Total Cost: €216.00
Daily
€7.20
Yearly
€2,592.00
Cost Breakdown
Power Draw (kW):0.45 kW
Hours per Month:600h
Average Load:80%
Electricity Rate:-€0.00
Your Monthly Cost:€216.00
Single PC Setup
Scenario
One gaming PC with a single GPU. Want to run AI locally for personal use.
Solution
Install rbee on your gaming PC. Run 7B-13B models locally with RTX 3070 or better.
Outcome
Personal AI assistant, code completion, document processing. Zero cloud costs, complete privacy.
Single GPUPersonal UsePrivacy
Multi-Node Homelab
Scenario
Multiple machines with GPUs scattered across your home. Want to run larger models.
Solution
Connect gaming PC + workstation + old server with rbee. Automatic SSH orchestration.
Outcome
Run 70B+ models by splitting across machines. Automatic load balancing, unified control.
Multi-GPULarge ModelsDistributed
Hybrid Setup
Scenario
Need privacy for sensitive workloads but want cloud burst capacity.
Solution
Mix local homelab GPUs with rented cloud GPUs. Privacy-first routing with rbee.
Outcome
Best of both worlds. Local for privacy, cloud for burst. Cost-optimized workload distribution.
HybridPrivacyCloud Burst
Security & Privacy First
Built for homelabbers who care about security and privacy.
process-isolation
Each model runs in isolated process. Clean shutdown, no orphans.
ssh-security
Uses your existing SSH keys and known_hosts. No new attack surface.
zero-telemetry
No phone-home. No tracking. Your data stays on your network.
open-source
GPL-3.0-or-later. Fully auditable. No proprietary blobs.
Process Isolation
Each worker runs in its own isolated process.
- Clean shutdown on exit
- No orphaned processes
- VRAM cleanup guaranteed
Network Security
Your homelab, your rules. LAN-only mode available.
- LAN-only mode (no internet required)
- SSH-only communication (port 22)
- VPN compatible (WireGuard, Tailscale)
Homelab FAQ
Getting Started
Features
Hardware
Troubleshooting
Ready to Get Started?
Turn Your Homelab Into an AI Powerhouse
Download rbee and start running LLMs across all your machines. Free, open-source, and built for homelabbers.
Works with your existing hardware • SSH-based orchestration • GPL-3.0-or-later