PC for OpenClaw Locally: Which Hardware to Choose? | Radiance Systems
Share
OpenClaw is an AI agent framework designed to run entirely on your machine. But to harness its full power — fast inference, heavy models, zero-latency pipeline — your hardware matters as much as your software setup. Here’s everything you need to know.
By the Radiance Systems team · · 8 min read
- OpenClaw runs 100% offline once models are downloaded — zero cloud dependency.
- The minimum viable for smooth use: an NVIDIA GPU with at least 16 GB of VRAM.
- For 30B+ models, aim for a GPU with 24–32 GB of VRAM or a multi-GPU system.
- System RAM: at least 32 GB DDR5 for mixed AI + office workloads.
- Radiance CoreAI workstations come preconfigured with CUDA, Ollama, and the environment ready for OpenClaw.
- No subscription, no cost per token: your AI belongs to you.
What is OpenClaw and why run it locally?
OpenClaw is an open-source AI agent orchestration framework. It allows you to deploy, chain, and interact with language models (LLM) directly from your infrastructure — without going through a third-party API, without sending your data to remote servers.
The promise is simple: you keep full control over your AI pipeline. Your documents, your queries, your responses never leave your network. For professionals subject to GDPR, medical confidentiality, or legal ethics, this is a fundamental paradigm shift.
But OpenClaw is resource-hungry. It relies on LLM models loaded into GPU memory, and its performance directly depends on your hardware. A consumer PC or laptop won’t hold up for long.
What PC configuration for OpenClaw locally?
OpenClaw orchestrates agents that call LLM models for inference. These models dictate the hardware requirements. Here are the thresholds to know:
The GPU: the key component
VRAM (video memory) determines which models you can load and how fast they generate tokens. It’s the number one limiting factor.
NVIDIA GPUs are preferred thanks to the CUDA ecosystem, essential for frameworks like llama.cpp, Ollama, or vLLM that power OpenClaw. The RTX 5000 series (Blackwell architecture) currently offers the best performance/VRAM ratio available in workstations.
System RAM
RAM is used for long context, embeddings, and parallel pipelines. Below 32 GB DDR5, you risk bottlenecks on complex workflows.
Storage
LLM models are heavy: from 4 GB for a quantized 7B model to 80 GB+ for a full 70B. Plan for at least a 2 TB NVMe SSD if you intend to store multiple models simultaneously.
Step-by-step OpenClaw installation guide on your PC
Here is the complete process to get OpenClaw running on Windows 11 or Ubuntu. Radiance CoreAI workstations come with CUDA drivers and Ollama preinstalled, reducing this process to 3–4 steps.
Make sure your NVIDIA driver is up to date (version ≥ 525) and that the CUDA Toolkit is installed. Check with nvidia-smi in the terminal. On Radiance machines, this is already done.
OpenClaw relies on Ollama to load and serve models locally. Install it from ollama.com or via terminal: curl -fsSL https://ollama.com/install.sh | sh. On Windows, the GUI installer is sufficient.
From the terminal: ollama pull mistral for a 7B, or ollama pull qwen2.5:32b for a more powerful model. The download happens only once.
Clone the official OpenClaw repository, install the Python dependencies (venv recommended), then configure config.yaml to point to your local Ollama endpoint (http://localhost:11434).
Start Ollama in the background (ollama serve), then launch OpenClaw. Your first AI agent runs entirely on your machine — no data leaves your network.
Add a RAG layer by indexing your PDFs, Word files, or local databases. OpenClaw can query your internal documents in natural language — professional secrecy guaranteed.
Hardware comparison: which workstation for which OpenClaw use?
Not all PCs are equal when it comes to OpenClaw. This table summarizes key criteria according to usage profiles.
| Usage profile | Required VRAM | System RAM | Supported models | Recommended tier |
|---|---|---|---|---|
| Light individual use — 1 agent, occasional use | 16 GB | 16–32 GB | 7B–13B Mistral, LLaMA 3.1 |
CoreAI 16 |
| Regular professional — RAG, multi-docs, complex agents | 16–24 GB | 32 GB | 13B–30B quantized Qwen 2.5, DeepSeek |
CoreAI 32 |
| Multi-user office — simultaneous inference | 32 GB | 64 GB | 70B quantized Mixtral, Qwen 72B |
CoreAI 64 |
| Multi-agent production 24/7 — fine-tuning | 64–96 GB (multi-GPU) | 128 GB | 70B+ full / 200B Fine-tuning possible |
Rack 2×5090 / GB10 |
Note on quantization: 4-bit or 8-bit GGUF models can run a 30B on 16 GB VRAM, with slight quality loss. For critical professional use, prefer maximum precision — which means properly sizing your GPU.
Radiance machines ready for OpenClaw
Each machine is assembled in Auriol (13), pre-configured and delivered ready to use. CUDA drivers, Ollama, and Python environment are already installed.
Which LLM models to use with OpenClaw locally?
OpenClaw is compatible with any model served via an OpenAI-compatible API — including Ollama, llama.cpp, LM Studio, or vLLM. Our 2025 selection:
| Model | Size | VRAM required | Strengths | Target profile |
|---|---|---|---|---|
| Mistral Small 3.1 | 24B | ~14 GB (Q4) | Fast, multilingual, precise instructions | General use, assistants |
| Qwen 2.5 / Qwen3 | 7B–72B | 4–40 GB | Excellent in French, strong reasoning | Legal, medical, accounting |
| DeepSeek-R1 | 7B–70B | 4–38 GB | Chain reasoning, code, analysis | R&D, engineering offices |
| LLaMA 3.3 | 70B | ~38 GB (Q4) | Reference model, versatile | All professional uses |
| Gemma 3 | 9B–27B | 5–16 GB | Lightweight, multimodal, performs well on small GPUs | Individual offices, medical |
For professional use in French, Qwen 2.5 32B and Mistral Small 3.1 offer the best balance between response quality and inference speed on a 16–24 GB GPU.
Local AI vs cloud: decisive arguments for professionals
| Criteria | Local AI (OpenClaw + Radiance) | Cloud AI (ChatGPT, Copilot…) |
|---|---|---|
| Data confidentiality | ✓ Total — nothing leaves | ✗ Data transmitted to provider |
| GDPR compliance | ✓ Native — zero transfer outside the EU | ⚠ Variable depending on DPA contracts |
| Cost model | One-time investment | Subscription + token billing |
| Offline availability | ✓ Works offline | ✗ Requires a permanent connection |
| Customization / fine-tuning | ✓ Total on your data | Limited depending on the offer |
| Professional secrecy | ✓ Respected by design | ⚠ Real ethical risk |
Radiance feedback: our pro tips for OpenClaw
We assemble and deploy AI workstations from Auriol for professionals across Europe. Here’s what our field feedback has taught us:
💡 Don’t underestimate NVMe bandwidth
At startup, OpenClaw loads the model from disk into VRAM. An NVMe Gen 4 (~7,000 MB/s) reduces this loading time by 40 to 70% compared to a SATA SSD. On our CoreAI machines, Gen 4 is the default.
💡 Limit the context window if you lack VRAM
OpenClaw can open long contexts (32K, 128K tokens). Each context token consumes VRAM. On a 16 GB GPU, limit to 8K–16K tokens to keep generation fast. On a 32 GB GPU, 128K tokens are accessible without issue.
💡 Enable Flash Attention if your GPU supports it
The RTX 5000 (Blackwell) natively supports Flash Attention 3, which reduces the memory footprint of long contexts by 30 to 50%. Enable it in your inference backend configuration.
💡 For a multi-user office: serve OpenClaw as an internal API
Rather than installing OpenClaw on each workstation, deploy a server instance on a Radiance rack workstation accessible via your local network. All your collaborators connect locally, data stays on-site, and you share GPU power.
Frequently asked questions about OpenClaw locally
Is OpenClaw difficult to install?
On a standard machine, installation requires terminal and Python knowledge. On Radiance CoreAI workstations, the environment (CUDA, Ollama, Python) is pre-configured, reducing installation to about ten minutes. Our team can also perform the initial setup remotely.
Can fine-tuning be done locally with OpenClaw?
OpenClaw is an agent orchestrator, not a fine-tuning tool. To fine-tune a model on your data, you will need a dedicated tool (Axolotl, LLaMA-Factory) and a GPU with at least 24 GB of VRAM for 7B models, or a multi-GPU system for larger ones. Our Rack 2×RTX 5090 and RTX 6000 PRO Blackwell configurations are sized for this use.
Does OpenClaw run on Windows?
Yes, OpenClaw supports Windows via WSL2 or natively with Python. Our CoreAI machines come with Windows 11 Pro and WSL2 pre-activated, ensuring full compatibility with the open-source AI tool ecosystem.
What is the difference between OpenClaw and Open WebUI?
Open WebUI is a graphical interface to interact with models via chat. OpenClaw is an AI agent framework — it automates complex tasks, chains model calls, integrates external tools, and can reason in multiple steps. Both can coexist on the same machine.
My firm has strict GDPR obligations. Is local AI really compliant?
Yes. As long as personal data processing takes place on your equipment, on your premises, without transfer to a third-party provider, you comply with the GDPR principles of data minimization and sovereignty. Radiance workstations are specifically designed to guarantee this architecture by default.
Your PC for OpenClaw locally, assembled in Auriol
Tell us your usage, your profession, your target models. We configure the ideal machine and send it to you ready to use within 4 to 10 business days.
Request a free quote →
