Watsonx Assistant for Z now runs with Spyre Accelerator, bringing generative and agentic AI to the IBM Z platform, designed to offer a trusted and secured, deployment option for customers who demand data privacy.
This enhancement marks the next evolution of IBM’s strategy to deliver secured, agentic AI and automation directly on IBM Z, helping enterprises run AI agents powered by large language models (LLMs) natively on-prem, within the trusted IBM Z environment, with full data privacy, governance and control.
As organizations adopt AI systems powered by LLMs, data privacy becomes mission critical. This is especially true for watsonx Assistant for Z, where AI agents work directly with sensitive operational data—system metrics, logs, configuration details and workflow inputs—to interpret intent and execute tasks. IBM Z environments handle customer records, financial transactions and regulated workloads that cannot risk exposure. As these AI agents become increasingly capable, ensuring that data remains protected and contained becomes even more critical.
In agentic AI systems like watsonx Assistant for Z, AI agents access and act on operational data to perform real work, from analyzing system performance to automating complex workflows. This autonomy requires an architecture that helps accelerate intelligence while data remains private, compliant and under enterprise control. IBM Z provides that foundation.
Running AI inference directly on-platform helps support data residency, built to minimize the need to move sensitive information to external servers or public clouds. With pervasive encryption, isolation capabilities, and hardware-based security built into the platform, IBM Z helps enable organizations to scale AI while supporting stringent security and governance requirements.
With the addition of the Spyre Accelerator, powered by the IBM Telum® II processor, watsonx Assistant for Z can run agentic and generative AI workloads securely where data already resides. Spyre enables high-performance, on-platform inference, accelerating insights while ensuring sensitive operational information can remain within the protection of IBM Z.
The Spyre Accelerator, introduced with the IBM z17 and IBM LinuxONE 5, is purpose-built to handle large-language models and generative and agentic AI workloads securely within enterprise systems.
It brings enterprise-scale AI inferencing directly to IBM Z, enabling large language models and agents to run entirely on the mainframe. This integration allows organizations to scale generative and agentic AI while preserving privacy, governance and reliability.
With watsonx Assistant for Z now leveraging Spyre Accelerator, helps enterprises gain even greater flexibility to deploy large language models natively on IBM Z. This release introduces support for IBM Granite foundation models—starting with the Granite 3.3-8B-instruct model, tested and optimized to run on IBM Z with Spyre cards for z17 deployments.
At the same time, we continue to support Llama-based deployments for customers running their LLMs on x86 infrastructure, driving choice and adaptability across environments. This allows organizations to choose models that best align with their performance, compliance, and infrastructure strategy, while leveraging Spyre’s accelerated, on-platform execution for generative and agentic AI at scale.
When combined, watsonx Assistant for Z and Spyre enable:
Together, these advancements help watsonx Assistant for Z deliver faster, secured results with reduced operational complexity, all within the trusted IBM Z environment.
Spyre Accelerator support for watsonx Assistant for Z will be generally available beginning 12 December 2025. This milestone unites IBM’s innovation in AI hardware and software—combining the Spyre Accelerator, Telum II processor, and watsonx Assistant for Z to deliver secured, on-platform intelligence for enterprise workloads.
With the rapid expansion of Agent Builder capabilities, we are deprecating Assistant Builder in this release and plan to remove it entirely in the next release. We encourage users to transition to Agent Builder to take advantage of its enhanced functionality and future roadmap.
Experience the future of mainframe operations.
See the watsonx Assistant for Z product page
Read how IBM is reimagining work with agentic AI