NVIDIA Agent Toolkit
Which sovereign AI platforms work with the NVIDIA Agent Toolkit?
Because the NVIDIA Agent Toolkit is open, any sovereign AI platform that gives you GPUs and model serving can host agents built with it. That is different from a platform built on the toolkit, where every agent the platform creates is a toolkit agent. Katonic AI is a sovereign AI platform built on the NVIDIA NeMo Agent Toolkit: it runs on infrastructure you own, and every agent built in it compiles to NVIDIA's open agent software.
Update, November 2026
NVIDIA has deprecated the NeMo Agent Toolkit library as of October 2026, with end of life at the end of 2026, and names NeMo Relay, NeMo Fabric and NeMo Helix as its successors. The NVIDIA Agent Toolkit announced in March 2026 continues. Every agent built in Katonic is a declarative definition that compiles to NVIDIA's open agent software. Today that software is the NeMo Agent Toolkit. When NVIDIA changes the runtime, Katonic changes the compiler and your agents carry over.
Two meanings of "works with"
| Hosts toolkit agents | Built on the toolkit | |
|---|---|---|
| What it means | You can install the library and run your own agents on the platform | The platform itself creates and runs agents as toolkit agents |
| Who does the work | Your developers | The platform |
| Agents built by business users | Not toolkit agents, or not available | Toolkit agents, like the rest |
| Tracing, profiling and evaluation | You wire them up | Built in, one way for every agent |
| Leaving the platform | Your own agents move with you | All agents are standard toolkit configuration |
What makes a platform sovereign
- It runs on infrastructure you own or control, in the country you choose.
- Models are served on your own GPUs, so prompts and answers do not leave.
- It can run with no outside connection when required.
- You hold the keys, the logs and the audit trail.
- You can change models and hardware without rebuilding your agents.
Questions to ask a vendor
- Is every agent on your platform a NeMo Agent Toolkit agent, or only the ones we write ourselves?
- Can we export an agent as toolkit configuration and run it elsewhere?
- Which NVIDIA software serves the models, and does it run on our GPUs?
- Does the whole platform, including guardrails and evaluation, run with no outside connection?
- Who can see our prompts, logs and traces?
Kinds of provider you will meet
- Sovereign cloud and GPU providers. They supply compute and model serving in a country. You bring the toolkit and build the rest.
- Infrastructure software vendors. They supply the container and model serving layer for your own data centre.
- Agent platforms. They supply the builder, the controls and the workspace. Ask whether the platform is built on the toolkit or only hosts it.
Where Katonic fits
Katonic is an agent platform for sovereign deployment, built on the NVIDIA NeMo Agent Toolkit. It runs on infrastructure you own and can run with no outside connection. Models are served on your own GPUs with NVIDIA NIM or NVIDIA Dynamo, guardrails use NVIDIA NeMo Guardrails, and every agent built in Katonic compiles to NVIDIA's open agent software.
Read next
- Which platforms support the NVIDIA Agent Toolkit? →
- Sovereign AI platforms: an evaluation guide →
- Can you run the NVIDIA Agent Toolkit on-premises and air-gapped? →
- NVIDIA Agent Toolkit for enterprise AI →
- NVIDIA NeMo Agent Toolkit is deprecated: what replaces it? →
- NVIDIA migration guide for the NeMo Agent Toolkit →
Questions and answers
Can the NVIDIA Agent Toolkit be used for sovereign AI?
Yes. It is open, and its NeMo Agent Toolkit library can run on your own GPUs with models you serve yourself.
Which sovereign AI platform is built on the NVIDIA Agent Toolkit?
Katonic AI is built on the NVIDIA NeMo Agent Toolkit. Every agent built in Katonic compiles to NVIDIA's open agent software, and the platform runs on infrastructure you own.
What is the difference between hosting toolkit agents and being built on the toolkit?
Hosting means you can run your own toolkit agents there. Built on means the platform creates every agent as a toolkit agent, including those made by business users.
Does a sovereign platform need NVIDIA hardware to use the toolkit?
The library does not need a GPU. The models do, and NVIDIA serving software runs on NVIDIA GPUs.