Nvidia's new 'Blueprints' promise plug-and-play AI agents
Key Points
- Nvidia has introduced "Blueprints", pre-built instructions that include all the necessary components and configurations for developing AI agents capable of planning and executing complex tasks on their own.
- Blueprints have been used to create applications such as automated code documentation, voice AI agents, structured reporting, research assistance, and virtual AI assistants. Nvidia also provides blueprints for PDF-to-podcast conversion and video analysis.
- The new Nemotron models, available in three sizes for different use cases and optimized for agentic AI, serve as the technical foundation for text processing (Lama Nemotron) and image analysis (Cosmos Nemotron), but commercial use requires Nvidia's AI Enterprise software.
Nvidia has unveiled a new approach to building AI agents called "blueprints" - essentially pre-built templates that give developers everything they need to create autonomous AI systems without starting from scratch.
These blueprints come packaged as "launchables" - pre-configured development environments that teams can spin up instantly. The goal is simple: help developers build AI systems that can perform complex tasks on their own, without reinventing the wheel each time.
Five AI companies have already jumped on board with their own specialized blueprints. CrewAI created one for automated code documentation, Daily focused on language AI agents, LangChain developed templates for structured reports, LlamaIndex designed research tools for blog posts, and Weights & Biases contributed blueprints for virtual AI assistants.
Nvidia has added two blueprints of its own: one that turns PDFs into podcasts and another for video analysis. According to the company, the video blueprint can analyze footage 30 times faster than real-time playback.
There's a catch, though: developers need Nvidia AI Enterprise Software to use these blueprints. The software works with data centers from Dell, HPE, Lenovo, and Supermicro, plus cloud services from AWS, Google, Microsoft, and Oracle.
New Nemotron models based on Llama and optimized for "agentic" AI
Under the hood, these blueprints run on Nvidia's new Nemotron models, built on Meta's Llama technology. The family includes two main members: Llama Nemotron for text processing and Cosmos Nemotron for image analysis.
Nvidia says they've specifically optimized these Llama-based models for AI agents, enhancing their ability to follow instructions, engage in conversation, make function calls, and handle programming and math tasks.
For developers who want to try it out, Nvidia offers two paths: members of the Nvidia Developer Program will soon be able to access the models through hosted APIs, or they can download them directly from build.nvidia.com and Hugging Face for "testing and research."
Business users can get the complete package - including NeMo, NeMo Retriever, and both Nemotron models as NIM microservices - through Nvidia's AI Enterprise platform mentioned earlier.
AI News Without the Hype – Curated by Humans
As a THE DECODER subscriber, you get ad-free reading, our weekly AI newsletter, the exclusive "AI Radar" Frontier Report 6× per year, access to comments, and our complete archive.
Subscribe now