Launched this week
Hopper
First agentic development environment for mainframe/COBOL
77 followers
First agentic development environment for mainframe/COBOL
77 followers
Hopper is the first agentic development environment for the mainframe. It's Cursor for mainframes. It combines a real TN3270 terminal, mainframe-aware panels for datasets, jobs, members, and spool output, and an AI agent that can operate across z/OS workflows. The agent can inspect datasets, read and edit PDS members, write JCL, submit jobs, parse JES output, explain failures, and help developers debug mainframe workflows faster. Hopper is available on Windows, Linux, and macOS.
Hopper
Hey Product Hunt 👋
Today we’re launching Hopper, the first agentic development environment for mainframes.
Mainframes running on COBOL are the 60-year-old computing platforms that still quietly run much of the modern economy: banks, payments, insurance, airlines, government systems, and more.
But they were built for expert humans using terminal screens, function keys, batch jobs, datasets, and highly specific workflows, not AI agents.
Modern AI coding tools assume GitHub, shells, files, package managers, and test runners. Mainframes are a completely different computing paradigm.
Hopper combines a real mainframe terminal, context panels for datasets and jobs, and an AI agent that can safely operate across mainframe workflows.
Our goal is simple: bring AI agents to the legacy systems that still run the world without pretending they are modern codebases.
You can request access to a mainframe on our page, and start playing with Hopper to see what an agentic mainframe environment feels like.
Would love to hear any feedback or comments!
Agentic AI is hitting the legacy systems. Bridging agentic AI to z/OS is genuinely hard—mainframes don't have Git/files/shells.
Curious to know how you are managing the AI agent’s memory and context while debugging. On mainframes, logs and job outputs can become huge, and the agent also has to remember the sequence of ISPF screens, multiple PDS members, and earlier debugging steps.
Hopper
@sinchana_v Appreciate the comment! You’re exactly right that z/OS breaks a lot of the assumptions agent systems usually lean on. Hopper is built to learn shop-specific conventions for the connected LPAR like compile procs, system symbols, dataset layout, LE runtime quirks and more. It is designed to understand and navigate ISPF/JES workflows.
For very large outputs, we try to keep the model focused on the relevant slices while still preserving the underlying artifact for the user to inspect. Hopper also keeps track of recent context, like submitted job IDs, edited members, and learned JCL conventions, so follow-up questions like “why did my last job fail?” have something concrete attached to it.
We’re still early and there’s a lot more we want to improve in the coming months! These are exactly the kinds of hard problems Hopper is meant to grow into.
@kevin_huang17 All the best on tackling this! Kudos to the team.
Genuinely didn't expect agentic AI to hit mainframes in 2026. The hard part isn't just terminal access — COBOL business logic is often undocumented and lives in people's heads. How does Hopper handle cases where the agent needs context that's nowhere in the codebase?
Hopper
@deepak_roshan_a We have a complementary platform called HyperTwin that conducts AI-driven interviews to extract exactly that type of institutional and tribal knowledge that lives in people's heads. https://www.hypercubic.ai/hypertwin
Hopper, HyperTwin and our other suite of mainframe products all work in harmony with each other.
I'm studying AI and we barely even touch COBOL. Does Hopper have any way to help the agent understand undocumented business logic, or does it only work with what's explicitly in the code and job outputs? Feels like that gap could make or break it for real enterprise adoption
Hopper
@shwet_gaur Great question, in the mainframe ecosystem, a vast majority of the rules and engineering conventions are undocumented, they live in people's heads.
Mentioned this in another comment but we also extract the tribal knowledge (https://www.hypercubic.ai/hypertwin) from people in addition to codebase understanding. All of this becomes a grounding context layer for the agent to perform valuable work on your mainframe stack.
Never thought I would see a COBOL AI agent, this is awesome!
Hopper
@kn0wn Thanks! :)