Forty years of data the new tools can't see
If you run an IBM i, you already know the shape of this problem. The system works. It has been working since before most analytics vendors existed. It holds orders, inventory, ledger, and the operational history of the business — and every modern tool that promises to answer questions about that data starts by asking you to move it somewhere else first.
That's not a small ask. It means a pipeline off the i, a warehouse to land it in, someone to own both, and a security review for the whole arrangement. By the time anyone asks their first question, a quarter has gone by and the data is a day old.
Meanwhile the questions don't stop. Which orders shipped short last week. Which accounts have open balances past terms. Whether the number in the ledger matches the number in the other system. Today those questions go to whoever knows the schema, and they go one at a time.
What already exists, and where it stops
There is a real market here, and it's thinner than you'd expect. As of our August 2026 landscape review:
IBM Db2 Web Query is being replaced by IBM Database Expert Query for i, announced in February 2026. It's browser-based reporting. Capable, and not conversational.
LANSA BI and aXes offer guided natural-language query in plain English — but it's pre-LLM guided query, driven from a browser, and it works within the queries it was set up to handle.
MCP servers for Db2 for i exist — several of them — exposing IBM i data to Claude, Cursor, and other AI IDEs, some with schema annotations that map six-character field names to something readable.
Those MCP servers are the interesting signal. They prove the demand is real, and they prove nobody has productized it. They're plumbing for a developer's AI editor, not something a plant manager opens on a phone.
(This section reflects publicly documented products as of 2026-08-20 and is worth re-checking before you quote it — this market moves.)
How GenViz connects
You install a desktop app on a machine that is already inside your network and already allowed to talk to the i. Then you point it at the system with an ODBC connection string — the same shape your other tools use:
DRIVER=IBM i Access ODBC Driver;SYSTEM=host;UID=user;PWD=pass;That's the whole setup. There is nothing to install on the i, no ETL job, no VPC peering, and no always-on service to get through a security review. Your files stay on the i — the only thing that moves is the answer to the question you asked.
That workstation needs the IBM i Access ODBC driver — the standard IBM i Access Client Solutions package. Most shops that already query the i from Windows or a Mac have it.
Or skip the workstation entirely
If you'd rather not depend on what's installed on anyone's laptop, run the same connection through a connector — a small service you deploy inside your own network and point at the i.
The connector image already ships the IBM i Access ODBC driver. Nothing is installed on anyone's machine, and nothing is installed on the i. You deploy one container, and every person in the workspace can ask questions of the system whether or not they have a driver, a VPN, or the right laptop.
That's also the answer for scheduled runs, for people working from home, and for phones — a laptop has to be awake and on the network to answer a question; a connector doesn't.
It is still your infrastructure. The connector runs in your network, holds your credentials, and talks to your i. We never get access to the system itself.
To be exact about what does move: when an answer becomes an extract, that Parquet file is stored in our cloud storage so it can be opened on another device and shared. Your library files are not copied — the result set is.
Queries are read-only, structurally
Write-shaped statements — INSERT, UPDATE, DELETE, DROP, ALTER, TRUNCATE, GRANT, CALL — are refused before they reach the driver. Not a permission you configure and
hope nobody changes: the query is rejected on the way out.
You should still connect with a read-only profile. But the tool doesn't rely on you having done it.
About those field names
RPG-era schemas are their own problem. Six-character column names, files that haven't been renamed since the eighties, and business meaning that lives in someone's head rather than in the catalog. Any tool that dumps a raw schema into a prompt and hopes for the best will struggle here, and we'd rather say so than pretend otherwise.
Two things make it work anyway.
GenViz reads views as first-class objects. If you've already built SQL views over your physical files — the ones with readable column names that your reporting tools use — it sees them, knows they're views, and prefers them. If you haven't, this is the highest-leverage afternoon your DBA can spend: a handful of views over the files people actually ask about turns an unreadable catalog into something anyone can query in plain language. The definitions stay in your database, under your control, always current. Nothing to sync, and no vendor modelling format to learn.
That matters more on the i than anywhere else, for a practical reason: a large library list contains far more objects than any model can hold in context at once. Curated views aren't a nicety here — they're what makes a big system answerable at all.
And it remembers what you tell it. Where it guesses wrong about a field, you correct it in the conversation, and the correction is stored for the environment so the next person's question benefits. Those notes are treated as hints — the authoritative definitions stay in your views, where the database enforces them.
It isn't magic and we won't describe it that way. But it's a long way better than handing
someone a SQL prompt against ORDHDR.
Db2 on other platforms
GenViz also connects to IBM Db2 LUW — the Linux, Unix, and Windows edition — through a connector service you host inside your network. Same questions, same shareable answers, different deployment shape, because those databases typically sit somewhere a laptop can't reach.
What you get back
A table. Something you can sort, check against the source, and send to the person who asked. Larger results become an extract you can download as Parquet and open in whatever tool you already use.
Answers share to a phone, which for this audience is often the actual requirement — someone on a floor needs a number that currently lives behind a green screen and a person who knows how to query it.
Frequently asked
Do I need to move data off the i? No bulk move — GenViz queries Db2 for i directly, with no pipeline and no warehouse. Individual answers are a different matter: when one becomes an extract, that result set is stored as a Parquet file in our cloud so it can be opened elsewhere and shared.
Does anything get installed on the IBM i itself? No — nothing, in either setup.
Do I need the IBM i Access ODBC driver? Only if you query from the desktop app directly. That workstation needs the driver. If you run through a connector instead, the driver already ships in the connector image and nobody's machine needs anything.
Can it change data? No. Write-shaped statements are rejected before the driver sees them. Connect with a read-only profile as well.
How does it find my files — does it use my library list? No. It reads the
SQL catalog (QSYS2.SYSTABLES), so it sees schemas directly rather than
depending on a job's library list. System libraries are skipped, and both physical and
logical files show up alongside SQL tables and views. Your profile's authority still governs
what it can read.
One honest limit: a multi-member physical file is addressed through SQL, which works against a single member. If you rely heavily on multi-member files, tell us — it is a solvable problem and we would rather hear about it early.
What about Db2 on Linux or Windows? Supported, through a customer-hosted connector rather than directly from the desktop.
Is this just an MCP server? No — though GenViz does ship one, so your coding agents can reach the same data. The difference is that this is an application with accounts, shared answers, and a phone client, rather than plumbing for a developer's editor.
Ask your i a question this week.
GenViz is in early access. Join the waitlist to get set up — or download the app for macOS, Windows, or Linux and have it ready when your access lands.