Red Hat introduces new platform to scale enterprise AI agents
Companies are beginning to accelerate the transition from AI pilots to production operations.
Red Hat introduced new capabilities designed to scale AI agents across enterprise hybrid environments.
Companies are facing new challenges as they move autonomous AI systems from experimental testing into production environments. In that context, Red Hat introduced Red Hat AI 3.4, a platform focused on deploying and managing agentic models and workflows across hybrid infrastructures.
The company said many organizations are shifting from being “token consumers” to “token providers” in order to better manage costs and enable private and sovereign AI use cases.
In that environment, Red Hat identified friction between developers and infrastructure administrators, along with the rise of “shadow AI” practices that introduce unmanaged risks and unpredictable costs, as some of the industry’s main challenges.
This marks another step toward enterprise AI platforms designed to scale autonomous agents with governance, traceability and operational control capabilities across cloud and infrastructure environments.
Architecture for enterprise AI.
Red Hat said the transition from experimental chatbots to production-grade autonomous systems requires a new operational and infrastructure approach. Under that strategy, the company is promoting a “metal-to-agent” architecture designed to combine scalable inference, operational control and autonomous agent deployment across enterprise environments.
The update incorporates tools for distributed inference, model management, AgentOps, observability, model evaluation and multilayer security, with the goal of providing a unified infrastructure for developers and IT operators.
The company said the expansion of AI agents is rapidly increasing inference demand, especially in enterprise environments where agents operate with greater autonomy and require enhanced traceability and control capabilities.
Red Hat AI 3.4 capabilities.
Governed Model-as-a-Service (MaaS):
This capability allows platform engineers to offer curated and validated models through API endpoints with enhanced security using standard interfaces compatible with OpenAI. This enables unified governance for both internal models and external APIs, integrated with identity provider-based authentication.
The MaaS capability is built on a high-performance distributed inference foundation powered by vLLM and llm-d, designed to maintain optimized and efficient model serving across infrastructure environments.
AgentOps:
Red Hat AI 3.4 incorporates capabilities to manage agents from development to production regardless of the framework being used. The tools integrate tracing, observability, cryptographic identity and lifecycle management for autonomous agents.
Integrated prompt management:
The platform provides unified tools to build and manage prompts as first-class data assets. Storing the instructions that drive models and agents in a centralized registry creates a single source of truth for both developers and administrators.
Automated evaluations for models and agents:
Red Hat AI 3.4 introduces the evaluation hub, a unified and framework-independent AI evaluation control plane for assessing large language models (LLMs), AI applications and agents. This replaces fragmented testing methods with a unified benchmarking approach designed to measure quality, accuracy, security and risk.
Multilayer security:
Automated adversarial scanning is integrated directly into the development lifecycle. Leveraging technologies from Chatterbox Labs and the Garak project, the platform analyzes risks such as jailbreaks, prompt injections and bias, combined with NVIDIA NeMo Guardrails for runtime security.
According to Red Hat, these capabilities are intended to provide a security-focused path for moving from experimental pilots to production-ready enterprise deployments.
Production-ready observability:
MLflow integration provides visibility into agent execution through OpenTelemetry, enabling end-to-end tracing of LLM calls, reasoning steps, model responses, tool execution and token usage. This creates a transparent audit trail for prompts, embeddings and RAG configurations throughout the lifecycle of models and agents.
MLflow also provides integrated experiment tracking and artifact management for generative and predictive AI use cases.
Identity-based governance:
Through SPIFFE/SPIRE, Red Hat AI enables organizations to replace hardcoded static keys with short-lived tokens. This supports least-privilege operations for autonomous agents and helps link agentic actions to verified identities.
Automated experiences:
Tools such as AutoRAG and AutoML automate tasks related to data retrieval strategy selection, predictive model building and optimization of traditional AI processes.
Hardware flexibility and managed clouds:
The platform extends its unified architecture to deployments across CoreWeave, Microsoft Azure and IBM Cloud, while also expanding Red Hat AI Inference beyond OpenShift to additional Kubernetes services in order to maintain operational consistency across hardware and cloud providers.
Red Hat AI Inference also introduces request prioritization, allowing interactive and background traffic to share the same endpoint while prioritizing latency-sensitive requests under heavy workloads. The platform also adds support for speculative decoding, a technology designed to improve response speeds by two to three times with minimal impact on quality and interaction costs.
Scalability and control.
The company said one of the main challenges surrounding autonomous agents is the lack of visibility into their decision-making processes. In response, Red Hat AI 3.4 incorporates observability, auditing and traceability capabilities to track how agents execute actions and use tools within enterprise environments.
Joe Fernandes, Vice President and General Manager of the AI Business Unit at Red Hat, said:
“The agentic era represents an evolution of our platform: from running traditional applications to powering intelligent and autonomous systems. We are defining the open standard for how enterprises run AI. By providing a hardened, metal-to-agent foundation for AI inference, MaaS and AgentOps, Red Hat delivers the operational assurance organizations need to innovate at scale while maintaining rigorous control.”
Beyond software capabilities, Red Hat AI 3.4 also adds day-one support for NVIDIA Blackwell and AMD MI325X architectures.
Urvashi Chowdhary, Vice President of Product Management for AI Services at CoreWeave, said:
“CoreWeave’s collaboration with Red Hat is built on a shared commitment to openness and delivering a high-performance inference foundation that enables enterprises to scale their most complex AI workloads. Long-running autonomous agents in the enterprise require a new level of infrastructure control and security to ensure reliable operations at scale.”
With Red Hat AI 3.4, the company aims to expand operational capabilities for organizations moving toward enterprise AI models centered on scalable inference, autonomous agent deployment and hybrid infrastructure governance.
|
|
