Questions

It does the same job, building and running MODFLOW models, with one difference that changes the workflow. What you build in the interface and the FloPy script are a single definition, so a choice made in one is already made in the other. There is no export step and no second version of the model to keep in sync.

It is also much younger than either of them, and under active development. There are MODFLOW packages it does not author yet. The Specifications page lists every one of them, package by package, so you can check yours before asking for access rather than after.

Standard MODFLOW. GWlink writes the same input files any other tool writes, and they open anywhere. Nothing you build here is locked to us.

Yes, for MODFLOW 6 and MODFLOW-USG. Every package is registered as either fully editable, importable with reduced editing, passed through untouched, or explicitly rejected with a reason. There is no fifth category where a package imports and quietly loses information, and that is by design rather than by luck.

The per-package matrix is on the Specifications page.

Nothing else. GWlink is a desktop application for Windows and Mac, and it brings its own Python runtime, so you do not need an existing Python environment and you do not have to install FloPy yourself.

The first time it starts it fetches the MODFLOW executables it runs, mf6, mfusg_gsi, triangle and gridgen, from the MODFLOW-ORG release. That happens once and it needs a connection; if you already have them, it uses yours.

Yes, and it is your own. GWlink runs the agent on your CLI account, and you choose the engine between Claude Code, Codex and Pi. We do not resell tokens and we take no margin on your usage, so what the agent costs you is what your provider charges, at their prices. Local models are not supported today.

After the first launch, the modelling does. Building a model, running it and reading the results all happen on your machine. Two things need a connection, the one-time download of the MODFLOW executables described above, and the agent, which talks to your provider every time you use it.

It changes the model the way you do, through the interface. Each step calls the same handler the on-screen control calls, GWlink navigates to it and applies the change while you watch, and the step is written into the script, which is the model definition. What the agent did is there to read, line by line.

It is a coding agent, so it can also read and write files in your project folder. The MODFLOW input files are not one of the things worth worrying about, because GWlink generates them from its own state and rewrites them on every write. An edit made straight into them is overwritten the next time the model is written, whoever made it.

Without the agent, nothing leaves. The model is built, stored and run locally.

The agent is an ordinary coding agent, running on your account, in your project folder, under your provider's terms. Whatever your organisation has already decided about coding assistants is the decision that applies here, because it is the same thing. And if the answer for a given project is no, you can build the entire model without ever opening it.

Thirty canonical models run in the validation suite, all thirty checked against their expected convergence, and twenty-one of them come from public and citable sources including the MODFLOW 6 examples, FloPy and GMDSI's Milford model. The Validation page shows the current state of each.

Beyond the suite, GWlink is the tool we use on our own consulting projects.

GWlink is in private beta, by invitation. Request access and we will get back to you about your project and what a licence would look like for it. There is no public price list while the beta is private.