NVIDIA shows how to run OpenClaw locally on RTX GPUs

OpenClaw (formerly known as Clawdbot and Moltbot) is an open-source, local-first AI agent designed to run continuously on your machine. Currently one of the fastest-growing repos on GitHub, it quietly handles background tasks across your emails, calendars, and files while learning your working habits.

What it actually does

Unlike traditional chatbots that sit idle until you prompt them, OpenClaw is designed as an always-on desktop assistant. It lives locally, monitors ongoing context across your local workspace, and expands its tool belt through modular skills.

Keeping an autonomous agent alive via commercial cloud APIs gets painfully expensive in a hurry, as constant reasoning loops burn through tokens 24/7. It also forces you to pump sensitive local documents into remote servers. Running the entire pipeline on your own machine completely eliminates API fees and locks your data down locally.

Getting it running on RTX hardware

NVIDIA's new deployment guide walks through hosting both OpenClaw and its driving language model locally on GeForce RTX, NVIDIA RTX GPUs, and DGX Spark workstations.

You can spin it up natively on Windows or use WSL for a streamlined Linux environment. On the inference side, it supports familiar local runtimes like LM Studio, Ollama, and vLLM. NVIDIA highlights models like Qwen 3.6 27B for RTX GPUs, leveraging Tensor Cores and CUDA acceleration to keep token throughput snappy.

Context window sizing is crucial for autonomous workflows. The guide stresses setting context to at least 32K tokens, and bumping it to 64K if your VRAM headroom allows. For heavy duty setups, NVIDIA highlights DGX Spark with its 128GB memory footprint, capable of running an optimized Qwen 3.6 35B checkpoint continuously without breaking a sweat.

Once the local server is alive, you simply paste your OpenClaw gateway URL and token into the browser, with the ability to switch models on the fly straight from the chat interface.

Why it matters

For power users sitting on decent RTX silicon, running OpenClaw locally turns consumer gaming cards into personal operations centers without the meter running. It proves that capable desktop agency does not inherently belong behind a subscription paywall.

Still, running an autonomous agent locally shifts the entire security burden directly to you. OpenClaw interacts with actual files and desktop applications, meaning untrusted third-party skills can create real operational hazards. We recommend keeping skills to a bare minimum and testing in isolated folders before granting access to your daily life.