主なポイント
- NVIDIA is releasing NeMo Switchyard, an open model-routing technology for agentic AI. NeMo Switchyard is designed to dynamically route AI workloads across models based on the needs of the task, while supporting an open and extensible routing ecosystem. NVIDIA is making NeMo Switchyard open source, enabling developers to build and contribute routing approaches for their own use cases.
- Agentic AI will be multi-model. Different tasks require different levels of reasoning, specialization, speed, and cost, making dynamic model selection increasingly important.
- Open ecosystems matter. Boomi believes enterprises should have choice across models, runtimes, and routing technologies, with routing becoming an important foundation for efficiency and richer multi-model planning and reasoning.
Enterprise AI will not converge on a single model; organizations will need the flexibility to use the right model for each workload.
No single model will be best at every task an AI agent performs. One step may require deep reasoning, another classification, and another a smaller specialized model optimized for speed or cost. That changes the question from “Which model should I use?” to “Which model is right for this task, right now?”
NVIDIA NeMo Switchyard is an important step in that direction. It uses model capability, cost, latency, infrastructure, and application context to inform routing decisions, providing an open routing layer that connects workloads with the models best suited to handle them.
Routing Becomes Part of the AI Architecture
As enterprises move toward more autonomous agents, routing becomes increasingly important.
An agent may make many model calls as it plans, reasons, uses tools, and evaluates results. Sending every step to the largest model can add unnecessary cost and latency, while relying only on smaller models can limit quality on harder tasks.
Routing offers a more flexible approach: match each task to the model that best fits it.
Over time, routing can become part of how AI systems plan and reason, selecting different models for different stages of work and escalating when greater capability is required.
Open Ecosystems Matter
One model, one runtime, or one routing technology will not necessarily remain the right choice for every workload. That’s why it’s important to be able to easily make changes when needed.
NeMo Switchyard reflects this philosophy through a provider-agnostic architecture that separates routing from the underlying model endpoint and translates between provider API formats so applications keep their native interface. It is also open source, allowing developers to create and contribute routing approaches.
This open approach aligns strongly with Boomi’s view. Just as enterprises operate across diverse applications, clouds, APIs, and data environments, their AI environments will also be heterogeneous. We expect to see, and drive, continued innovation across models, runtimes, and routing technologies. What matters is preserving enterprise choice.
Agentic AI will be a multi-model world, where the right model is dynamically selected for the right task. NVIDIA NeMo Switchyard is an exciting step forward, bringing open, flexible model routing to agentic workloads. At Boomi, we see routing as key to improving AI cost and latency, while also creating the foundation for richer multi-model planning and reasoning. Just as Boomi simplifies complex hybrid landscapes, we believe the future of AI should embrace an open ecosystem of models, runtimes and routing technologies.
The future of enterprise AI will be defined by how intelligently organizations bring many models together, while preserving choice, control, and openness.
Learn more about NVIDIA NeMo Switchyard in NVIDIA’s blogs: