Tools We Recommend
Everything you need to run OpenClaw. We've tested all of these. Affiliate links help support this site.
Hardware
Apple Mac mini M4 (16GB)
Best dedicated OpenClaw host. Silent, always-on, low power. Handles models up to 14B.
EVGA GeForce RTX 3090 (24GB)
Best value local-LLM GPU. 24 GB VRAM runs Qwen 27B-class at Q4 (~35 tok/sec) — often half a 4090's price used.
GIGABYTE GeForce RTX 4090 (24GB)
Same 24 GB as the 3090 but faster (~50 tok/sec on Qwen 27B Q4). The pick if you're buying new and also game or render.
NVIDIA RTX PRO 6000 Blackwell (96GB)
Workstation tier. 96 GB VRAM runs 70B-plus at long context, or several models resident at once on one card.
Hosting (Cloud VPS)
DigitalOcean
$200 free credit with referral. Best for beginners. One-click Docker deploys.
AWS Lightsail
$5-20/mo. We have a full video tutorial for this. Great if you're already on AWS.
Hetzner
Cheapest cloud VPS in Europe. Popular with self-hosters. $4-10/mo for OpenClaw.
AI Models
Local LLM Fit and Speed Estimator
Estimate whether your model fits, where it bottlenecks, and whether to change RAM, VRAM, runtime, quantization, or context length. Open estimator.
Ollama (Free, Local)
Run models on your own hardware with zero API costs. See our benchmarks and recommendations.
Anthropic Claude API
Best tool-calling reliability. ~$6-13/mo for personal use. Pay-as-you-go API keys.
OpenRouter
One API key, access to 100+ models. Good for testing different models without separate accounts.
Guides
Model fit, expected bottleneck, runtime, quantization, and context-length triage.
macOS, Linux, Windows. 5-minute install.
Lock down your agent before connecting it to anything.
Gateway token, blank responses, Docker, macOS, Telegram, Windows.
Some links are affiliate links. We only recommend tools we actually use. Affiliate revenue supports this site.