Build on IBM watsonx and Red Hat

Agents for hybrid estates,
governed from the start.

watsonx Orchestrate, Granite, watsonx.data and Red Hat OpenShift, engineered into agents that work across cloud, data centre and mainframe, with governance built into delivery.

Talk about your project Explore the IBM stack

How we see it

Hybrid is not a phase. Design for it.

Many IBM estates run across public cloud, private data centres and IBM Z or Power systems, often under sovereignty rules. Moving everything first is rarely the right answer.

We build agents that reach data where it lives, run where policy allows and produce the governance evidence your risk team asks for, using watsonx and Red Hat OpenShift as the common layer.

Hybrid by design

Each part runs where it belongs.

One agent system can span IBM Cloud, your own data centre and IBM Z. Red Hat OpenShift keeps it consistent, and governance records every decision along the way.

Zone 1

IBM Cloud

watsonx Orchestrate
watsonx.ai and Granite
watsonx.governance

Managed services for building, running and governing agents.

Zone 2

Your data centre or another cloud

Red Hat OpenShift AI
Granite served with vLLM
watsonx.data

Models and data stay where your rules require.

Zone 3

IBM Z, LinuxONE and Power

Db2 and core transactions as tools
On-chip inference with Telum II
Spyre Accelerator

Agents reach the systems of record without moving them.

The governance evidence trail

Inventory
Every model, agent, prompt and dataset in use, with an owner.
Risk assessment
The use case, its impact and the controls it needs.
Evaluation
Test sets, scores and the thresholds a release must meet.
Approval
Who signed off, when and on which evidence.
Monitoring
Drift, failures and policy breaches in production, with alerts.

An illustrative design

An underwriting assistant across hybrid systems.

An underwriter asks for a summary of a renewal. The agent reads the policy file, checks the core system on the mainframe and drafts a recommendation. The underwriter decides, and the evidence is recorded.

#	Step	Where it runs	What is recorded
01	Underwriter asks	watsonx Orchestrate	Request and user identity
02	Agent plans the steps	Orchestrate ADK agent	Plan and agent version
03	Reads the policy file	watsonx.data · Docling	Documents retrieved
04	Checks the core system	MCP tool → Db2 on IBM Z	Tool call and response
05	Drafts the summary	AI Gateway · Granite or Claude	Model, prompt and output
06	Underwriter decides	Decision stays with a person	Decision and reviewer
07	Evidence recorded	watsonx.governance	Linked to the use-case record

The IBM AI stack

Six layers. We build across all of them.

Open a layer to see the IBM services involved and the part we take on.

ExperienceWhere people ask, review and approve.

watsonx Orchestrate chat
Slack and Teams
Business applications
IBM Bob for developers

What we build. We bring the agent into the tools people already use and design how it explains its reasoning and asks for approval.

AgentsWhere the plan is made and the steps are run.

watsonx Orchestrate
Agent Development Kit
Langflow
LangGraph agents
Agent Catalog

What we build. We build agents with the Orchestrate ADK, bring in agents from other frameworks and coordinate them in multi-agent workflows.

Tools and actionsWhere the agent touches your systems.

MCP tools
OpenAPI tools
ContextForge
IBM Confluent
IBM Event Streams

What we build. We expose core systems as MCP or OpenAPI tools, including those on IBM Z and IBM i, with narrow permissions and recovery paths.

Knowledge and dataWhere answers come from.

watsonx.data
Docling
watsonx.data integration
watsonx.data intelligence
Db2

What we build. We prepare documents with Docling, connect governed data through watsonx.data and keep lineage visible from source to answer.

ModelsWhere reasoning happens.

watsonx.ai
Granite 4
AI Gateway
Claude
vLLM on OpenShift AI

What we build. We test Granite, Claude and open models on your tasks and run them where your data rules allow, in IBM Cloud or on your own hardware.

Trust and operationsWhere you keep control.

watsonx.governance
Guardium AI Security
AgentOps
Instana
Red Hat OpenShift

What we build. We record models, risks and evaluations in watsonx.governance, trace agents with AgentOps and watch the platform with Instana.

What changed on IBM

What IBM shipped, in order.

The releases that change how we design and build on IBM, with a link to each announcement.

Oct 2025
Claude comes to IBM software
IBM and Anthropic announced Claude in select IBM software, starting with IBM’s AI development tool, and a guide to building secure agents with MCP.
Read the announcement (opens in a new tab)
Nov 2025
Red Hat AI 3
OpenShift AI 3 brought distributed inference with llm-d and vLLM, for serving models on your own infrastructure.
Read the announcement (opens in a new tab)
Apr 2026
IBM Bob is generally available
An AI development partner that routes work across Claude, Granite and other models. Mainframe, IBM i and Java modernization packages followed in July.
Read the announcement (opens in a new tab)
May 2026
IBM Sovereign Core
Sovereign operations built on OpenShift and Red Hat AI reached general availability at Think.
Read the announcement (opens in a new tab)
Aug 2026
Granite 4.2 adds reasoning
IBM’s open, Apache 2.0 models gained step-by-step reasoning while staying small enough to run on your own hardware.
Read the announcement (opens in a new tab)

Our services on this platform

From first idea to production on IBM.

How each stage of our work maps to IBM. Start at whichever stage fits where you are.

01Start

Decide what is worth doing.

Challenge24
An Orchestrate agent with Granite or Claude, on your examples, in three business days.
watsonx Orchestrate · watsonx.ai

02Build

Build the system.

Applied AI and agents
ADK agents connected to your core systems through MCP and OpenAPI tools.
Orchestrate ADK · MCP · watsonx.data
AI-infused software
Applications on OpenShift that run the same in IBM Cloud and your data centre.
Red Hat OpenShift · Db2 · Event Streams

03Strengthen

Make it dependable.

Enterprise AI Hardening
Guardrails, AgentOps tracing and AI security before broader rollout.
Guardium AI Security · AgentOps
Legacy Modernization
COBOL, PL/I and IBM i applications understood and modernized with their rules kept.
IBM Bob · Z and IBM i packages

04Run

Keep it healthy.

SRE as a Service
OpenShift and application telemetry triaged by agents, decided by engineers.
Instana · OpenShift
AI FinOps
Cloud and Kubernetes spend made visible and right-sized.
Cloudability · Turbonomic · Kubecost
AI Governance and Evaluation
Model inventories, risk assessments and evaluation records in one place.
watsonx.governance

05Extend

Add capacity.

Engineering teams
Engineers who know watsonx, OpenShift and Ansible, inside your team.
OpenShift · Ansible · Terraform

Claude on IBM

Claude in IBM software

Claude is built into IBM Bob and can be selected through the watsonx Orchestrate AI Gateway alongside Granite. We test both on your task and deployment constraints, and use Granite where a smaller open model running on your own infrastructure is the better fit.

How we build with Claude

Under the agents

The cloud foundation we build on.

Compute and containers

Red Hat OpenShift · IBM Kubernetes Service · Virtual Servers

One container platform across IBM Cloud, other clouds and your data centre.

Enterprise systems

IBM Z · LinuxONE · Power · IBM i

Connect agents to the systems that hold your most important records, without moving them first.

Data

Db2 · Cloudant · IBM Cloud Object Storage

Relational, document and object data, integrated or migrated with care.

Events and integration

IBM Confluent · IBM Event Streams

Event-driven connections with clear processing, retry and operational behaviour.

Automation

Ansible · Terraform · Schematics

Versioned infrastructure and repeatable environments across the hybrid estate.

Security and operations

IBM Cloud IAM · Key Protect · Instana · Sovereign Core

Access, encryption, observability and sovereignty controls in the architecture from day one.

Questions

Common questions.

When would you choose Granite models?

Granite is a strong option when you need open, Apache 2.0 models that are small enough to run on your own infrastructure, or when a task does not need a larger model. We test it alongside Claude and other options on your tasks.

Can watsonx Orchestrate work with agents built elsewhere?

Yes. Orchestrate can coordinate agents built with its own ADK, Langflow and frameworks such as LangGraph, and connect tools through MCP and OpenAPI.

Can you connect agents to our mainframe?

Yes. We expose selected transactions and data on IBM Z or IBM i as tools with narrow permissions, so agents can read or propose changes without the core system moving.

What does AI governance involve in a project?

Knowing which models, agents and data are in use, recording evaluations and risk decisions, defining access and monitoring behaviour. We set this up in watsonx.governance and agree the evidence your organization needs.

Can you work on-premises or under sovereignty rules?

Yes. With Red Hat OpenShift AI, vLLM and Granite, models can run on your own hardware. We assess options such as IBM Sovereign Core against your obligations.

Do we need to replace our data warehouse?

Usually not. We start from your existing systems and add retrieval or a complementary data layer with watsonx.data where it helps.

Can you help modernize COBOL or IBM i applications?

Yes. We use IBM Bob and its modernization packages to understand and document the code, then plan an incremental modernization that keeps the business rules intact.

Can you help operate Red Hat OpenShift?

Yes. We can scope platform, deployment and operational support around your environment, with responsibilities and response arrangements agreed up front.

Let’s start with the work you want to improve.

Tell us about your application, your IBM environment, or the idea you want to explore.

Talk about your project

Agents for hybrid estates,governed from the start.