Enterprise-Grade AI Inference, Driven Natively from z/OS — Where Your Data Lives
Your mainframe contains your most valuable business data and decades of proven logic. Moving it off-platform introduces risk, cost, and complexity.
Connect/AI is Data 21’s complete AI integration layer for z/OS — scheduled batch inference inside your existing JCL, and a browser-based agentic workspace for real-time work, against any model: mainframe-resident, your own local model, or a cloud provider like Claude, OpenAI, or Grok. It’s built on Connect/Web’s native z/OS HTTP foundation, and reaches into Connect/3270 to mine and act on live 3270 screen data — all without leaving the mainframe. Built on high-performance, z/OSMF-compatible access with no JVM layer, Connect/AI delivers a smaller footprint and lower overhead — all while fully leveraging native System-SSL and SAF security.
Connect/Web and Connect/3270 already ship with MCP API access and a chat-only workspace on their own — Connect/AI is what turns that connectivity into a full agentic stack: tool-use workspace and native batch agentic inference, from day one.
Connect/AI’s Complete AI Stack
Connect/AI reaches z/OS resources through Connect/Web’s native infrastructure, and mines live CICS, VTAM, and z/OS 3270 screen data through Connect/3270 — turning both into inference, agentic automation, and real-time decision support. With Connect/AI, you can:
All processing can remain securely on the mainframe, protected by native System-SSL encryption and SAF security controls — fully respecting data sovereignty and data gravity.
Supported Languages & Integration
Faster claims processing and exception handling
Intelligent operations and performance analysis
Automated policy reviews and customer service workflows
Compare Your Options
Connect/MP is designed to complement and extend IBM’s mainframe AI and modernization stack. While IBM provides powerful models, governance, and core services, Connect/MP delivers specialized native capabilities that help organizations operationalize AI more efficiently on z/OS.
Scroll to see all columns
| Feature / Aspect | Connect/AI (Complete Stack) | IBM Traditional Stack (z/OSMF) | IBM watsonx.ai (on Z) | Cloud AI + Connector | Custom / Open-Source |
|---|---|---|---|---|---|
| Native z/OS Execution | Excellent — fully native binaries | Good (JVM-based) | Partial | No | Varies |
| Batch JCL Integration | Excellent | Limited | Good | Limited | Manual |
| 3270 Navigation & Agentic Flow | Best-in-class — native structured screen mining + full agentic loop, approval-gated for writes | Limited | Limited / requires custom integration | Limited (scraping) | Custom |
| Agentic Capabilities | Very strong — built-in reasoning + approval-gated 3270 actions | Limited | Strong (general agents) | Good | Limited |
| Web AI Agentic Workspace | Yes — real-time, browser-based agentic AI console (native on z/OS) | No native equivalent | Yes (watsonx studio, cloud/hybrid) | Yes (cloud consoles) | Varies |
| Performance & Efficiency | Superior: tens of MB footprint, no JVM/GC, low latency | Higher overhead | Higher overhead | Network bound | Varies |
| Security & Compliance | Excellent — native SAF throughout | Strong | Strong | Weaker | Depends |
| Data Movement | None for on-Z operations | Minimal | Minimal on-Z | Required | None (if on-Z) |
| Zero App Code Changes | Yes — target application unchanged | Often required | Often required | High effort | High effort |
| Pricing Model | MIPS-based, no toll fees | Included in z/OS + Zowe | Subscription + consumption | Consumption | Free + infra |
| Deployment Effort | Low | Medium–High | Medium | High | Very high |
| Skills Required | Low–Medium: e.g., JCL, REXX, basic prompt engineering | Medium: JCL + web services / Java | High: prompt engineering, AI/ML, integration | High | Very high: AI engineering + mainframe dev |
| Best For | High-performance native agentic AI on files and live 3270 data with an interactive workspace | General management & modernization | Broad enterprise AI | Hybrid/cloud | Custom control |
Comparison reflects typical deployment patterns as of 2026 and general industry positioning; specific results vary by environment and configuration.
Connect/AI · Complete AI Stack · Data 21
Contact us today for a personalized demonstration of Connect/AI’s complete AI stack.
Secure · Native · Intelligent · Sovereign · High-Performance
Data Sovereignty by Design. Built for the Mainframe.
