Factory Enables Autonomous Software Engineering for High-Security Environments with NVIDIA Nemotron

Factory is bringing the autonomous software factory to DGX Spark, built with NVIDIA Nemotron 3.5 Lightning.

For security-sensitive teams, software work cannot leave their environment. Source code, operational context, and execution must stay within their control. Factory gives teams a way to build and evolve software locally with DGX Spark and Nemotron 3.5 Lightning, without treating local control as an exception.

The software factory with total, secure control

Factory undertakes comprehensive engineering tasks, including understanding issues, modifying code, running tests, and preparing changes for review. Teams define the intent while the software factory executes the output.

DGX Spark brings that execution close to the code. It is suited to legacy modernization, critical infrastructure, financial services, government, and public-sector environments where the workload is sustained and the data is sensitive.

Local control needs open weights

For on-device deployments, teams need a model they can run, govern, and customize inside their own network. Open-weight models are foundational when deploying locally. Factory can connect to an approved local deployment of NVIDIA Nemotron 3.5 Lightning or another open-source model on DGX Spark.

NVIDIA Nemotron 3.5 Lightning and other open-source models give Factory a customizable model option for software-engineering workflows. Factory's collaboration with NVIDIA assists with reproducible software-engineering workflows that help evaluate and improve open models.

Nemotron 3.5 Lightning powers always-on agents

Always-on agents handle complex, multi-step work autonomously: researching, deciding, and acting without constant human input. They gather context, observe their environment, reason over available information, and take action, with every step invoking a language model.

But every step does not require the same model. Frontier models excel at complex reasoning and orchestration, while specialized models are better suited to high-volume, repetitive, domain-specific tasks. Effective agentic AI relies on a system of models that routes each task to the model best equipped to handle it.

NVIDIA Nemotron 3.5 Lightning is the fastest open model in its class for always-on agents and specialized workloads. Built as a 30B MoE model with 3B active parameters, it delivers up to 4x higher throughput and enables 30% faster task completion across high-volume agentic workflows. Learn more here.

What Factory x NVIDIA DGX Spark enables

• A local alternative to data center compute
• An autonomous software factory that operates near sensitive code and infrastructure
• A deployment path for regulated, sovereign, and high-security environments
• Support for NVIDIA Nemotron 3.5 Lightning and other open-source models that enterprises can run on their own terms

Example deployments

A regulated bank can run NVIDIA Nemotron 3.5 Lightning on DGX Spark inside its own environment and connect to Factory. When a production issue appears in a payment, risk, or reporting system, Factory investigates the relevant code, modifies the implementation, runs tests, and prepares a change for engineer approval. The bank retains control of the DGX Spark deployment, model access, and release process.

A government organization can deploy DGX Spark in an environment approved for sensitive operational information, running NVIDIA Nemotron 3.5 Lightning or another approved open-source model. Factory can then help modernize a legacy application, resolve a defect, or build an internal workflow without moving the model workload outside that environment. For air-gapped deployments, Factory can support the appropriate installation path.

Getting started

Get started by signing up for Factory and following the Quickstart guide.

For on-premises and air-gapped deployment, contact sales@factory.ai. Learn more about Factory for enterprise.