Autonomous, self-evolving agents can reason, act, and persist across sessions without constant human input. They generate code, create sub-agents, and execute complex workflows, but running them safely requires both control and reliable model endpoints.
Developers are increasingly moving from experiments to production-ready workflows, spinning up long-running agents that rely on multiple open-source models to handle coding, planning, and tool use autonomously.
Lightning + NVIDIA OpenShell: Safe and Scalable Agent Execution
NVIDIA OpenShell is an open-source runtime designed to run autonomous agents safely. It sits between the agent and the underlying infrastructure, controlling what the agent can see and do, where inference runs, and enforcing policies at the infrastructure level. NVIDIA OpenShell runs agents in isolated sandboxes with zero code changes, starting them with no permissions and keeping all inference private by default.
Through our partnership, Lightning integrates with NVIDIA OpenShell by providing high-performance frontier open model endpoints, such as Kimi K2.5, Minimax M2.5, and GLM-5. Developers authenticate with Lightning inside OpenShell, enabling agents to leverage our compute and models safely, efficiently, and at scale.
This combination delivers:
- Reliable endpoints for long-running agents: Run complex agent workflows without worrying about infrastructure or scaling.
- Access to high-quality open models: Frontier open-source models that rival closed alternatives, optimized for cost and performance.
- Infrastructure-level safety: Fine-grained controls and policy enforcement keep agent actions private and secure.
Why It Matters
Lightning AI and NVIDIA are also working together on NVIDIA NemoClaw, an open source stack that simplifies running OpenClaw always-on assistants, more safely, with a single command. As part of the NVIDIA Agent Toolkit, it installs the NVIDIA OpenShell runtime, a secure environment for running autonomous agents, and open source models like NVIDIA Nemotron.
Autonomous agents can now execute multi-step workflows that previously required human supervision. With Lightning providing high-performance endpoints for Kimi K2.5, Minimax M2.5, and GLM-5, developers can run coding, planning, and tool-use tasks directly inside NVIDIA OpenShell. This lets teams focus on designing and testing agent workflows, integrating them into applications, and scaling them efficiently, without worrying about infrastructure or model orchestration.
