PC for Local OpenClaw: What Hardware to Choose? | Radiance Systems
Share
OpenClaw is an AI agent framework designed to run entirely on your machine. But to get the most out of it — fast inference, large models, low-latency pipelines — your hardware matters as much as your software configuration. Here’s everything you need to know.
By the Radiance Systems Team · · 8 min read
- OpenClaw works 100% offline once the models are downloaded — zero cloud dependencies.
- The minimum viable setup 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 CoreIA workstations come preconfigured with CUDA (ROCm on the CoreIA Radeon AI PRO R9700), Ollama, and an 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 (LLMs) directly from your own infrastructure — without using a third-party API or sending your data to remote servers.
The promise is simple: you maintain complete control over your AI pipeline. Your documents, queries, and 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-intensive. It relies on LLMs loaded into GPU memory, and its performance depends directly on your hardware. A consumer PC or laptop won't hold up for long.
What PC configuration do you need for local OpenClaw?
OpenClaw orchestrates agents that call LLMs for inference. These models dictate the hardware requirements. Here are the thresholds you need to know:
The GPU: the centerpiece
VRAM (video memory) determines which models you can load and how fast they generate tokens. It is the number one limiting factor.
NVIDIA GPUs are preferred thanks to the CUDA ecosystem, which is best supported by the frameworks that power OpenClaw, such as llama.cpp, Ollama, or vLLM. On the R9700, Ollama runs via ROCm. The RTX 5000 series (Blackwell architecture) currently offers the best performance-to-VRAM ratio available in workstations.
System RAM
RAM is used for long contexts, embeddings, and parallel pipelines. With less than 32 GB of DDR5, you may experience bottlenecks in complex workflows.
Storage
LLMs are large: from 4 GB for a quantized 7B model to 80 GB+ for a full 70B model. Plan for at least a 2 TB NVMe SSD if you intend to store multiple models simultaneously.
Step-by-step guide to installing OpenClaw on your PC
Here is the complete process for getting OpenClaw up and running on Windows 11 or Ubuntu. Radiance CoreIA workstations come with CUDA drivers (ROCm on the CoreIA Radeon AI PRO R9700) and Ollama preinstalled, reducing this process to 3–4 steps. If you prefer to receive a local AI machine already ready for OpenClaw, we can configure it before shipping.
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 the 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 only needs to be done once.
Clone the official OpenClaw repository, install the Python dependencies (a virtual environment is recommended), then configure config.yaml to point to your local Ollama endpoint (http://localhost:11434).
Start Ollama in the background (ollama serve), then run OpenClaw. Your first AI agent runs entirely on your machine — no data leaves your network.
Add a RAG layer by indexing your PDFs, Word documents, or local databases. OpenClaw can query your internal documents in natural language — professional confidentiality guaranteed.
Hardware comparison: which workstation for which OpenClaw use case?
Not all PCs are equal when it comes to OpenClaw. This table summarizes the key criteria according to usage profiles.
| Usage profile | VRAM required | System RAM | Supported models | Recommended tier |
|---|---|---|---|---|
| Light individual use — 1 agent, occasional use | 16 GB | 16–32 GB | 7B–13B Mistral, LLaMA 3.1 |
CoreIA RTX 5060 Ti |
| Regular professional use — RAG, multi-docs, complex agents | 16–24 GB | 32 GB | Quantized 13B–30B Qwen 2.5, DeepSeek |
CoreIA Radeon AI PRO R9700 |
| Multi-user office — simultaneous inference | 32 GB | 64 GB | 8-bit 30B Mixtral, Qwen 32B |
CoreIA RTX 5090 |
| Multiple agents on a large model — fine-tuning | 64 GB (multi-GPU) / 128 GB unified | 128 GB | 8-bit 60B or 4-bit 70B (2x RTX 5090) / 4-bit 200B (GB10) Fine-tuning possible |
CoreIA 2x RTX 5090 / GB10 |
Note on quantization: 4-bit or 8-bit GGUF models can run a 30B model on 16 GB of VRAM, with a slight loss in quality. For critical professional use, opt for maximum precision — which means properly sizing your GPU.
Radiance machines ready for OpenClaw
Each machine is assembled in Auriol (13), preconfigured, and delivered ready to use. CUDA drivers (ROCm on the CoreIA Radeon AI PRO R9700), Ollama, and the Python environment are already in place.
Which LLMs should you use with local OpenClaw?
OpenClaw is compatible with any model served via an OpenAI-compatible API — this includes 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-of-thought reasoning, code, analysis | R&D, design offices |
| LLaMA 3.3 | 70B | ~38 GB (Q4) | Reference model, versatile | All professional uses |
| Gemma 3 | 9B–27B | 5–16 GB | Lightweight, multimodal, high performance on small GPUs | Individual practices, 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: the deciding factors for professionals
| Criterion | Local AI (OpenClaw + Radiance) | Cloud AI (ChatGPT, Copilot…) |
|---|---|---|
| Data confidentiality | ✓ Total — nothing leaves | ✗ Data transmitted to the provider |
| GDPR compliance | ✓ Native — zero transfer outside the EU | ⚠ Varies by DPA contracts |
| Cost model | One-time investment | Subscription + per-token billing |
| Offline availability | ✓ Works without internet | ✗ Requires a permanent connection |
| Customization / fine-tuning | ✓ Total on your data | Limited by the offer |
| Professional secrecy | ✓ Respected by design | ⚠ Real ethical risk |
Radiance Lessons Learned: Our Pro Tips for OpenClaw
We assemble and deploy AI workstations from Auriol for professionals across Europe. Here’s what our field experience has taught us:
💡 Do Not Underestimate NVMe Bandwidth
At startup, OpenClaw loads the model from the disk into VRAM. A Gen 4 NVMe (~7,000 MB/s) reduces this loading time by 40% to 70% compared to a SATA SSD. On our CoreIA machines, it’s Gen 4 or Gen 5 NVMe.
💡 Limit the context window if you lack VRAM
OpenClaw can handle long contexts (32K, 128K tokens). Each context token consumes VRAM. On a 16 GB GPU, limit it to 8K–16K tokens to maintain fast generation. On a 32 GB GPU, 128K is easily accessible.
💡 Enable Flash Attention if your GPU supports it
RTX 5000 (Blackwell) GPUs natively support 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 firm: serve OpenClaw as an internal API
Rather than installing OpenClaw on every workstation, deploy a server instance on a Radiance rack workstation accessible via your local network. All your employees connect locally, data remains on-site, and you share the GPU power.
Frequently asked questions about local OpenClaw
Is OpenClaw difficult to install?
On a standard machine, installation requires basic knowledge of the terminal and Python. On Radiance CoreIA workstations, the environment (CUDA or ROCm on the R9700, Ollama, Python) is pre-configured, reducing installation to about ten minutes. Our team can also perform the initial setup remotely.
Can you perform local fine-tuning 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 models. Our CoreIA 2x RTX 5090 and CoreIA RTX PRO 6000 configurations are sized for this use.
Does OpenClaw run on Windows?
Yes, OpenClaw supports Windows via WSL2 or native Python. Our CoreIA workstations come with a choice of Windows 11 Pro with WSL2 pre-enabled or Ubuntu, ensuring full compatibility with the open-source AI tool ecosystem; the CoreIA GB10 comes with DGX OS, based on Ubuntu.
What is the difference between OpenClaw and Open WebUI?
Open WebUI is a graphical interface for interacting 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 the processing of personal data takes place on your hardware, on your premises, without transfer to a third-party provider, you are in compliance with the GDPR principles of data minimization and data sovereignty. Radiance workstations are specifically designed to ensure this architecture by default.
Your PC for local OpenClaw, assembled in Auriol
Tell us your use case, your profession, and your target models. We will configure the ideal machine and send it to you ready to use after a maximum of 15 business days for assembly.
Request a free quote →
