Hermes AI Agents Run Locally on NVIDIA RTX and DGX Spark
Caroline Bishop May 13, 2026 13:45
Hermes brings self-evolving AI agents to NVIDIA RTX PCs and DGX Spark, leveraging Qwen 3.6 models for unmatched local performance.
Hermes, a groundbreaking self-evolving AI agent developed by Nous Research, is now optimized for local use on NVIDIA RTX PCs, PRO workstations, and DGX Spark systems. Announced on May 13, 2026, Hermes leverages NVIDIA hardware and the latest Qwen 3.6 large language models (LLMs) to deliver high performance in autonomous workflows.
Since its release, Hermes has gained significant traction, crossing 140,000 GitHub stars in less than three months and becoming the most widely used agent according to OpenRouter. Designed to write and refine its own skills, Hermes stands out for its reliable, always-on performance, making it a preferred choice for developers and AI enthusiasts looking for robust local agent solutions.
What Makes Hermes Unique?
Hermes introduces several standout features that differentiate it from existing agent frameworks:
- Self-Evolving Skills: The agent learns and improves autonomously, refining its skills based on complex tasks and user feedback.
- Contained Sub-Agents: Tasks are segmented into short-lived sub-agents, minimizing confusion and optimizing resource allocation.
- Optimized Reliability: Every tool and plugin is rigorously tested by Nous Research, ensuring seamless operation even with large local models.
- Framework Superiority: Developer tests consistently show Hermes outperforms competing agents, thanks to its active orchestration layer.
These capabilities make Hermes ideal for 24/7 local deployment, with NVIDIA RTX GPUs providing the computational power to unlock its potential.
Qwen 3.6: A Leap in Local AI Performance
Hermes relies on the latest Qwen 3.6 models from Alibaba, which outperform prior-generation models with significantly smaller memory footprints. The Qwen 3.6 35B model matches the performance of older 120B-parameter models while requiring only 20GB of memory. Similarly, the 27B variant delivers accuracy comparable to 400B-parameter models at a fraction of the size.
These models, optimized for NVIDIA hardware, enable Hermes to handle complex tasks quickly and efficiently. NVIDIA Tensor Cores further enhance performance, reducing latency and increasing throughput for multi-step workflows and self-improvement tasks.
Why DGX Spark is the Ideal Host
For users seeking an all-in-one solution, NVIDIA DGX Spark offers unparalleled support for agentic AI. With 128GB of unified memory and 1 petaflop of AI performance, DGX Spark can sustain Hermes and other AI agents in continuous, high-demand environments. It is particularly suited for developers running multiple workloads simultaneously on advanced models like Qwen 3.6.
For those getting started, NVIDIA provides detailed playbooks and hands-on sessions through its "Build It Yourself" AI series. These resources guide users in deploying Hermes on DGX Spark, leveraging tools like LM Studio and Ollama for seamless integration.
Getting Started
Hermes is open-source and available on its GitHub repository. Paired with NVIDIA RTX GPUs or DGX Spark systems, it offers an accessible entry point for developers and AI enthusiasts eager to explore the frontier of local autonomous agents.
As the demand for self-evolving AI grows, Hermes’ combination of adaptability, reliability, and local-first design makes it a key player in the next wave of AI innovation.
Image source: Shutterstock.jpg)