NVIDIA brings agents to life with DGX Spark and Reachy Mini
These articles are AI-generated summaries. Please check the original sources for full details.
NVIDIA brings agents to life with DGX Spark and Reachy Mini
Today at CES 2026, NVIDIA demonstrated a system integrating the DGX Spark with the Reachy Mini robot, showcasing a functional AI agent capable of interacting with the physical world. The system leverages NVIDIA’s Nemotron reasoning LLMs and Isaac GR00T N1.6 open reasoning VLA, alongside the NVIDIA Cosmos world foundation models.
Why This Matters
Current AI assistants often operate in isolated digital environments, lacking the ability to meaningfully interact with the physical world. While ideal models promise seamless integration, the reality is complex and resource-intensive; deploying these models requires substantial compute and storage. This setup, requiring ~93GB of disk space for models, highlights the practical challenges of bringing advanced AI capabilities to a personal robotics platform, addressing a critical gap in current AI assistant technology.
Key Insights
- DGX Spark & Reachy Mini Demo: NVIDIA showcased a working agent at CES 2026.
- Composable Architecture: The system utilizes a modular approach, combining open models for reasoning and vision with an agent framework for orchestration.
- NeMo Agent Toolkit: NVIDIA’s NeMo Agent Toolkit enables flexible configuration and experimentation with different models and behaviors, and is used by developers to build agentic workflows.
Working Example
# Clone the repository
git clone git@github.com/brevdev/reachy-personal-assistant
cd reachy-personal-assistant
# Set up environment variables (example)
# Create a .env file with:
# NVIDIA_API_KEY=your_nvidia_api_key_here
# ELEVENLABS_API_KEY=your_elevenlabs_api_key_here
# Start the NeMo Agent Toolkit service
cd nat
uv venv
uv sync
uv run --env-file ../.env nat serve --config_file src/ces_tutorial/config.yml --port 8001
Practical Applications
- Personal Robotics: Enables users to create personalized AI assistants with physical embodiment for tasks like information retrieval and object manipulation.
- Developer Tooling: Provides a platform for developers to experiment with and build advanced agentic systems, leveraging open-source components and NVIDIA hardware.
References:
Continue reading
Next article
NVIDIA Cosmos Reason 2 Brings Advanced Reasoning To Physical AI
Related Content
NVIDIA Cosmos Reason 2 Brings Advanced Reasoning To Physical AI
NVIDIA released Cosmos Reason 2, a vision language model achieving #1 open model status on the Physical AI Bench and Physical Reasoning leaderboards.
OpenMind OM1: Building an Open Source Operating System for Humanoid Robots
Jan Liphardt introduces OM1, an open-source robotic OS that leverages large language models for data fusion and utilizes $1,250 hardware components with 10,000-hour durability to enable human-centric robot interactions, shifting the focus from complex motor tasks like onion chopping to social engagement and spatial understanding.
NVIDIA Isaac Enables Healthcare Robot Development with SO-ARM Starter Workflow
NVIDIA Isaac's SO-ARM starter workflow facilitates the development and deployment of autonomous medical robots by streamlining data collection, training, and evaluation using simulation and real-world hardware.