Klique adds NVIDIA Nemotron 3.5 Super VL, Dynamo, OpenShell support

Klique positions itself as an enterprise control plane for AI resource management, governance, and cost visibility, and this announcement extends its existing collaboration with NVIDIA across three operational areas: compute cost and data sovereignty, production-scale agents, and securing autonomous workflows. By supporting Nemotron 3.5 Super VL, Klique lets users run NVIDIA's open model on their own hardware, which the company says ensures no data leaves the tenant and removes third-party per-token billing. The addition of NeMo Switchyard's routing algorithms alongside Klique's LLM gateway gives organizations a way to balance spending on frontier models against strict data sovereignty needs.

On the agent side, the body notes that a single agent task can trigger many model calls as it plans, uses tools, and reasons through each step, so inference demand climbs quickly once agents reach production. NVIDIA Dynamo is built for that load, distributing inference across GPU workers and sending repeated requests to workers that already hold the relevant cached context. Klique customers can deploy Dynamo directly from the platform and run agents on their own GPU infrastructure or in the cloud, with Klique adding security access rights, observability, and LLM routing to divert large-scale workflows to cost-effective resources. Teams can also retrain agents on past results, with the platform recording the data, model version, and compute cost behind every update so changes can be reviewed and traced.

For security, NVIDIA OpenShell establishes a monitored runtime environment designed to protect internal files, networks, and credentials during agent execution. By integrating OpenShell, Klique says enterprises can deploy autonomous agents like those built with NemoClaw open blueprints, Claude Code, Pi, Hermes, and Codex with the governance layer, policy compliance, and audit trails required for production.