Work on mainframes and COBOL

Tools

custom-built or unnamed agent

Overview

The first agentic development environment for the mainframe.

Workflow

The first agentic development environment for the mainframe. Use AI agents to navigate TN3270, inspect datasets, write JCL, debug jobs, query VSAM, and operate inside z/OS from a modern development environment. An agent that understands z/OS Drives ISPF by panel ID, writes column-strict JCL, decodes spool failures into structured diagnostics, and queries VSAM as SQL. Compile, test, ship in one prompt Drives the JCL, parses JES return codes, NEWCOPYs into CICS, and pauses for your approval before every change. Debug failed jobs without hunting through SDSF Hours of SDSF triage in one @-tag. The agent decodes JESMSGLG, JESYSMSG, and SYSUDUMP into the abend code, failing step, and source line. A real TN3270 terminal, when you need it Works exactly like the TN3270 you've used for years. Full PF, PA, and attention-key support. Pricing Hobby Includes: No credit card required macOS, Windows, and Linux Connect to your own mainframe All core Hopper features Enterprise Everything in Hobby, plus: SAML SSO MCP Server Access Admin and Model Controls Org-wide Privacy Controls No Model Training Priority Support On-prem / VPC Deployment SOC 2 Type II report available under NDA Download Hopper, connect to your LPAR, and start shipping macOS Apple Silicon.dmg · arm64 Intel.dmg · x64 Windows Installer.exe · x64 Linux .debDebian, Ubuntu · x64 .AppImageAny distro · x64 Frequently Asked Questions Still have questions? What is Hopper? Hopper is a desktop application for z/OS engineers that bundles a TN3270 terminal, a datasets browser, a jobs browser and an AI agent into a single interface. In natural language, you can ask the AI Agent to read COBOL source, submit and monitor batch jobs, decode VSAM records and write changes under your supervision. Hopper runs on macOS, Windows and Linux. What operating systems does Hopper run on? Hopper runs on macOS, Windows and Linux. Hopper provides separate builds for Apple Silicon and Intel processors on macOS. On Windows, Hopper provides an x64 installer. On Linux, Hopper supports Debian and Ubuntu via a .deb package and any Linux distribution via an AppImage. Does Hopper require anything to be installed on the mainframe? No. Hopper runs entirely on your laptop and requires no mainframe-side installation. Nothing runs on z/OS, so your sysprog does not need to deploy anything. Hopper connects to your mainframe over standard TN3270 and optional FTP protocols using your existing network connection. How does Hopper connect to z/OS? Hopper connects to z/OS using two standard mainframe protocols: TN3270 for the live terminal session and FTP for datasets and the JES queue. TN3270 is required. FTP is optional but recommended; without FTP credentials, the datasets browser, jobs browser, builds and most agent tools stay disabled. What can Hopper's AI agent do on the mainframe? Hopper's AI agent reads COBOL and JCL source, submits and monitors jobs, decodes VSAM records and drives ISPF using plain English prompts. The agent operates across six tool categories spanning datasets, jobs, terminal, CICS, COBOL and data taps, and can chain work across subsystems in a single conversation turn. Hopper's datasets tools list, read, write and allocate sequential datasets, PDSes and members. Hopper's jobs tools submit JCL, wait for completion, read spool output and classify failures. Hopper's terminal tools drive the TN3270 session, read screens and send keystrokes and attention keys. CICS operations in Hopper include sign-on, transaction execution, resource installation and CEDA defines. COBOL support in Hopper covers compilation, copybook parsing and VSAM record decoding. Hopper's data tap tools snapshot VSAM data into a queryable local copy and run SQL against it. What mainframe languages does Hopper understand? Hopper's AI agent handles COBOL, JCL, REXX and CLIST. The agent traces how programs and datasets connect and works out what to edit for the task you describe. The agent reads the source directly from your PDSes and uses it within the current conversation. How does Hopper's approval model protect against unintended mainframe changes? Hopper's approval model separates read-only actions from mutating actions. Read-only actions such as listing members, reading source or querying spool run instantly without confirmation. Mutating actions such as writing a member, submitting a job or installing a CICS resource pause in chat for your explicit approval before anything on the mainframe changes.

Reported result

The source does not provide measured outcomes, efficiency gains, or specific success metrics for this workflow. It describes the capability to reduce SDSF triage time and enable 'compile, test, ship in one prompt' but lacks quantitative data. Proposed acceptance check: Measure the reduction in time spent manually navigating ISPF panels and decoding JES return codes for a standard batch job failure compared to traditional methods.

Human role

The original post does not say how a person stayed involved.

Limitations

Prerequisites include a valid z/OS LPAR connection, TN3270 access, and recommended FTP credentials for full functionality. Without FTP, datasets and jobs browsers remain disabled. Enterprise features like SAML SSO and on-prem deployment require custom pricing. Source reliability is secondary (vendor marketing). Human approval is mandatory for all write operations to prevent unintended mainframe changes.

Try it

Act as a mainframe developer using Hopper. My goal is to debug a failed batch job and update a COBOL program. Inputs: - Mainframe LPAR: [replaceable_lpar_name] - Job Name: [replaceable_job_name] - Dataset: [replaceable_pds_name] - Error Context: [replaceable_error_description] Steps: 1. Connect to the LPAR via TN3270 and FTP. 2. Use the agent to retrieve the spool output for [replaceable_job_name] and decode the abend code and failing step. 3. Read the source code from [replaceable_pds_name] using the agent. 4. Propose a fix for the COBOL code based on the error. 5. STOP and wait for my explicit approval before writing the member or submitting the job. 6. Once approved, submit the job and monitor for completion. Deliverables: - A structured diagnostic report of the original failure. - The proposed code diff. - Confirmation of job submission status. Quality Check: Ensure the agent does not execute any mutating commands without my 'approve' command in the chat. If FTP credentials are missing, report which tools are disabled. If the abend code is unknown, ask for clarification rather than guessing.

More cases · Software & development

• Update a code base with Copilot chat and agent mode
• Check code against user intent with review agents and tests
• Build a budgeting app from a single prompt
• Build a mental models framework with a multi-agent dev team
• Build a instant-opening markdown viewer for macOS