OL Code · Agentic Mainframe Development
Mainframe vibe coding
Sounds like an oxymoron. It isn't. OL Code gives an AI agent what a senior mainframe developer already has: secure host access, the application graph, the data, the tests, and the documented intent. Your team keeps the review, the pipeline, and an easy way back.
The hard part was never the AI
Seven problems we solved before an agent could touch COBOL on z/OS
01 / SETUP
One workspace, set up once
A browser IDE with central administration: host, connection, compile templates, CICS settings, and the AI provider and keys, configured once for every developer. No local installs, no sources on laptops.
02 / CONTEXT
The graph beside the editor
The open member shows its documentation, callers, copybooks, transactions, DB2 tables, VSAM files, and the business operations it belongs to. The agent reasons over the estate, not one file.
03 / ACCESS
Reach the host as it is
Connect through the RSE API or JZOS, on premises, inside the platform, or as a lite bundle beside the host. Every action runs as the developer's own RACF user. No shared service account.
04 / TESTING
Testing inside the loop
Compile and follow the job, refresh the CICS program, run a test asset, and compare runs before and after. Wrap a program as a callable asset and run JSON test cases against it.
05 / DATA
Test data one click from the code
Run an EXEC SQL block as a read-only DB2 query, or page through the decoded VSAM records behind a CICS READ, even before you save. Same sign-in, same security.
06 / KNOWLEDGE
Specs from the open program
Generate a governed specification with diagrams from the business operations the graph found, right beside the source. Shared context for the reviewer, the agent, and whoever comes next.
07 / VIBE CODING
A harness built for mainframe work
Describe the change and the agent makes it in the workspace, with mainframe skills for COBOL columns, BMS maps, clean compile, and RPC extraction. You review the diff before anything reaches the host.
08 / SHIPPING
Git, CI/CD, and one-click rollback
Branch, commit, and push from the IDE in USS, follow Endevor promotion through its Git bridge, or restore any saved version of a PDS member.
Video / AI coding agent
The agent edits, compiles, and repairs COBOL
Working beside the host explorer and editor, the OL Code agent searches the program library, applies a targeted source change, creates a new COBOL member, and saves the updates to the host. It then diagnoses an intentional compile failure, repairs the source from the compiler listing, and confirms a clean compile.
The agent harness
Every turn runs in a mode that decides what the agent may do
ASK
Read and search
Answers questions about programs and the estate. Never changes code.
PLAN
Propose first
Lays out a larger change for you to approve before anything is edited.
AGENT
Make the change
Edits the workspace copy. Accepting the diff is what saves the member to the host.
CLEAN COMPILE
Fix until it builds
Edits and recompiles until the listing is clean.
Video / Context, data, and specs
From a CICS screen to a reviewable spec
Starting from a CICS transaction, the related COBOL opens in the editor and the AI assistant explains how the programs work together. AtomIQ analysis adds documentation, operations, business objects, and SQL details, the embedded SQL jumps straight to query results, and the session ends with a generated OL Spec covering business capability, entities, user journey, and process flow.
Bring your own agent
The same host tools, over MCP
Prefer Claude, Cursor, or your own agent? OL Code exposes its host tools over MCP: search, read, compile, jobs, CICS logs, and test assets. They run in the signed-in user's host context, and each turn gets only the tools its mode allows.
Developer in charge
No tool writes straight to the host
The agent edits the workspace copy, you review the diff, and only accepting it saves the member. The agent asks before a test run unless you have given it full access, and data access from the editor is read-only by design: only SELECT and WITH reach DB2, and only read commands reach VSAM.
Shipping it
Mainframe code joins your CI/CD
In the USS work mode, OL Code clones the team repository into a Unix workspace, and developers branch, commit, pull, rebase, and push from the sidebar. Your existing pipeline takes over from there. Running Endevor? Keep it as the system of record, and let its Git bridge carry workspace commits through Endevor-governed promotion.
Easy way back
Rollback for PDS members, one click
A PDS member has no version control of its own, so OL Code keeps one for you. Every save keeps a copy labeled by where it came from: the editor, the agent, or the host text captured just before the save. Compare any copy side by side and restore it. Nothing touches the host until you save.
Video / Modernize in the same loop
Turn a green-screen program into a callable service
Vibe coding is not only for fixes. Starting from a green-screen account update in OL Terminal, the screen COBOL opens in OL Code, the conversion wizard resolves its copybooks, and the program becomes RPC COBOL that keeps the business rules. The new member is saved, compiled, and tested with JSON input and output, ready to publish as an API.
At a glance
Built so teams can trust the result
Agent modes
Ask, Plan, Agent, and Clean compile decide what each turn may do.
Direct agent writes to the host
Every change goes through a diff you accept.
Local installs
One centrally administered browser workspace for the whole team.
Click to restore
Any saved version of a PDS member, compared side by side first.
ECOSYSTEM
Works with the mainframe you already run
Native host access and your existing security, source control, and pipeline.
FAQ
Common questions
Does source code leave the mainframe?
No. Source stays on the host and developers work on it from the browser. When you start analysis from the explorer, only dataset and member names are sent. The workspace still connects to your Git and CI/CD when you want it to.
How is host security enforced?
Developers sign in with their own host credentials, and every browse, read, edit, compile, and submit runs in their own host context. RACF decides what they can touch, exactly as it does today.
Can the AI agent change production code on its own?
No. The agent edits a workspace copy, you review the diff, and only accepting it saves the member to the host. Plan mode adds an approval step before larger changes start.
Can we use our own AI agent?
Yes. OL Code's host tools are available over MCP, running in the signed-in user's host context with the same per-mode tool limits.
How do changes reach our pipeline?
Through Git in the USS work mode, through Endevor's Git bridge if Endevor is your system of record, and with built-in version history and restore for PDS members.
Try mainframe vibe coding on your own code
Tell us about your z/OS environment and the change you want an agent to make, and we will set up a working session.
Thanks — we received your message.
Something went wrong. Please try again.