Nutanix Enterprise AI 2.8

Nutanix Enterprise AI 2.8 gives you more control over AI access, expands agent and model capabilities, and makes fine-tuning easier to manage. The release also adds more deployment options for NVIDIA NIMs and new capabilities for operating AI at scale.

New

• Control access with greater precision. Assign predefined or custom roles, control individual actions, and limit users to specific resources or resources they own.
• Share individual models securely. Give specific users or groups access to selected imported models without exposing your entire model catalog.
• License Agent Gateway capabilities separately. Add and manage an Agent Gateway license for access to unified endpoints and Model Context Protocol (MCP) servers.
• Ground Agent responses in your own documents. Use trusted documents to generate more relevant answers with citations that link back to the source content.
• Use fine-tuning with general availability. Customize supported models with more control over training parameters, resources, and output settings.
• Deploy NVIDIA NIMs in air-gapped environments. Import model files from NFS or S3-compatible storage and deploy them without requiring access to external registries.
• Import additional NVIDIA NIMs. Deploy supported NIMs beyond the pre-validated catalog using container images hosted in the NVIDIA NGC registry.
• Use MCP with general availability. Manage MCP client keys more easily and forward required request headers to MCP servers.
• Scale local endpoints across multiple GPU nodes. A new tech preview lets supported models use multiple GPU nodes with the vLLM inference engine.
• Offload KV cache to CPU memory. A new tech preview gives local endpoints another way to manage GPU resources by moving KV cache storage to CPU memory.
• Deploy more pre-validated models. Add support for new Mistral and NVIDIA Nemotron models, including Mistral Large 3 and NVIDIA Nemotron 3 Ultra.

Improvements

• View and download MCP server logs from the console. Monitor server activity and retrieve logs for troubleshooting and analysis without leaving Nutanix Enterprise AI.
• Use Anthropic Messages API with unified endpoints. Agent Gateway unified endpoints now support the Messages API specification for Anthropic provider APIs.
• Set more precise rate limits. Identify clients by API client key or request header and apply default or client-specific limits to unified endpoints.
• Manage fine-tuning jobs with more flexibility. Monitor training metrics, pause and resume jobs, and import completed fine-tuned models directly into Nutanix Enterprise AI.
• Customize local endpoint configurations. Add and validate custom vLLM arguments and environment variables, review parsed settings, and update parameters for hibernated experimental endpoints.

Known Issue

• Internal authorization policies may remain visible after a signed-in user revokes their own access to those policies.