Governed On-Prem AI Automation

Enterprise-grade AI agents are powerful, but for many regulated industries, they’re difficult to deploy in practice.

Most agent frameworks rely on cloud-hosted models, requiring sensitive enterprise data to be sent to external APIs. For financial services, healthcare, insurance, and government environments, that architecture can be limiting.

At Automation Anywhere, we’ve been evaluating a different approach, running advanced open-weight models within our automation platform, where universal orchestration connects models, enterprise systems, and governed workflows.

We recently benchmarked NVIDIA Nemotron 3 Super 120B to assess its ability to power enterprise-grade agentic automation fully on-prem.

The results are compelling.

Enterprise-grade performance, on-prem

Across 150+ enterprise agent scenarios, Nemotron 3 Super demonstrated strong performance on complex, multi-step workflows. Scenarios spanned insurance claims, financial operations, healthcare compliance, supply chain, procurement, and customer retention.

On our enterprise benchmark:

It delivered task completion performance comparable to leading cloud-based models.
It outperformed almost all large closed-source models we evaluated.
It is the best-performing open-source model we have evaluated at its size on our benchmark.

These findings indicate that open-weight models running entirely within enterprise infrastructure can now meet the performance bar required for serious agentic workloads, without relying on external APIs.

Governed agent reasoning inside the security perimeter

Using Nemotron 3 Super, we tested agent reasoning, planning, and structured tool use within orchestrated workflows.

Rather than deploying unconstrained autonomous agents, we embed AI agents inside governed processes orchestrated through our Mozart Orchestrator.

This enables:

Structured tool invocation
Deterministic workflow scaffolding
Human-in-the-loop checkpoints
Enterprise-grade controls and observability

Additionally, our execution model grounds AI execution in enterprise process expertise through Automation Anywhere’s Process Reasoning Engine (PRE), which provides structured reasoning across enterprise workflows.

Running this stack fully on-prem allows organizations to deploy agentic automation while maintaining compliance, data residency, and operational control.

Extending automation with vision (also on-prem)

Enterprise environments often include thick-client applications, legacy systems, and virtualized desktops, where selector-based automation can fail.

In addition to Nemotron 3 Super for reasoning, we are evaluating NVIDIA Nemotron vision-language models to enable vision-based UI automation, which allows the system to visually identify and interact with application controls when deterministic methods break.

This expands automation coverage across complex enterprise systems without introducing cloud dependencies.

A unified on-prem architecture

A single NVIDIA GPU can power:

Nemotron 3 Super 120B for reasoning
Nemotron 2 Nano VL vision-language models for video and image understanding
Automation Anywhere’s orchestration and governance layer

No external API calls.

No data leaving the building.

For regulated enterprises, this unlocks AI-powered automation in environments where cloud-based models aren’t viable.

Why this matters

AI agent discussions often focus on model capability. For enterprises, deployment architecture is just as important.

Our evaluation of NVIDIA’s latest Nemotron 3 Super release shows that high-performing, open-weight models can now support enterprise-grade agent workflows — while remaining fully on-prem.

That combination of performance, control, and simplicity is what makes AI-powered automation viable for the industries that need it most.

Learn more

Explore how Automation Anywhere’s APA System enables agentic automation across mission-critical processes in regulated industries, including financial services, healthcare, insurance, and the public sector.