FAQ

What is Qubis, exactly?

A native desktop application that puts an engineering agent, a code editor, a 3D viewport, and a local solver runtime in one window, working on files on your own machine.

How is it different from the web application?

The web application is a hosted CFD workspace you drive through a browser: you upload geometry, and meshing and solving run on Qubis's infrastructure.

Qubis is installed on your machine. Your files stay local, the agent works in the directory you attached, and solvers run as containers on your own CPU. The .nv* project files are the same in both, so a case moves between them.

Do my files get uploaded?

Your files live on your disk. The agent runs against Qubis's backend, so the content it reads in order to answer you is sent to the model as part of the conversation — that is how it can read a case file and tell you what is wrong with it.

Solves are local: Ferro runs as its own process beside Qubis, and third-party solvers run as containers on your machine through navier-solverd.

Does it work offline?

Partly, but you have to get in first. Launch is gated on a sign-in, which needs a connection, and a launch that cannot reach the sign-in service stops at the gate. Past it, the editor, viewers, viewport, Solver Desk, and local git all work without a connection; the agent does not, since it needs the backend.

What does it cost to run?

Usage is metered per organization and shown in Settings → Usage: tokens, calls, and cost. Effort settings are the biggest lever — a high-effort session costs materially more than a low-effort one.

Which platforms?

macOS on Apple Silicon, Windows on x86-64, and Linux on x86-64. See Install & Sign In.

Do I need Docker?

Only for third-party solvers. They run as containers, so they need a container engine. Ferro ships with Qubis and needs nothing. The viewers are all in process. See Environment Setup.

Which solvers can it run?

Ferro (structural FEM) is built in. Twelve third-party families ship in the runtime catalog: OpenFOAM, code_aster, SU2, Yade, CalculiX, OpenModelica, ngspice, Elmer, Basilisk, Project Chrono, openEMS, and Palace. You can add your own runtimes with a custom catalog. See Solver Desk.

Can I use my own build of a solver?

Yes. Point Settings → Solvers → Custom runtime catalog at a JSON file of runtime descriptors. An entry whose id matches a built-in one replaces it, so you can repoint a runtime without forking anything.

Will closing Qubis kill my solve?

No. navier-solverd is spawned detached, and Qubis adopts an already-running daemon on the next start rather than orphaning runs in flight. The daemon shuts itself down after an idle period with no runs and no requests.

Can the agent delete my files?

Yes, using shell commands within the workspace you attached, subject to your permission mode. A shell deletion or rename does show in the thread diff, which reads git, but nothing in Qubis undoes it. Review destructive commands in shell approval cards when prompted, and commit or back up your work before a permissive session. Bypass is the last mode in the ⇧Tab cycle rather than the first.

Can I stop it mid-task?

⌃C always interrupts. Esc does too if you enable it in Settings → Agent. Neither is an undo — files already written stay written.

How do I keep the agent off my working copy?

Use a worktree. On a new, empty thread in a git repository, toggle worktree in the composer and the thread works in its own checkout on its own branch.

Can I run several agents at once?

Yes — run several threads. On the same repository, give each thread its own worktree, otherwise they edit the same files.

What is the difference between a skill and an MCP server?

A skill is knowledge: it teaches the agent how to do something with tools it already has. An MCP server is capability: it gives the agent tools it did not have. A plugin is the folder that can carry either, or both. See Skills and Plugins & MCP.

Can I write my own skills?

Yes. Create one from New › Skill or drop a folder with a SKILL.md into your skills directory. A skill of yours that shares a name with a bundled one overrides it. See Skills.

Can I use my own MCP servers?

Yes. New › MCP writes one as a plugin under <config>/prism/plugins/. See Plugins & MCP.

Which model does it use?

Whichever you pick in the composer. Settings → Agent controls which models appear in the picker and which is the default.

Does it edit files behind my back?

It edits files it tells you about, and in a git repository every edit appears in the thread diff. In Accept Edits and Bypass it does not stop to ask first — which is exactly why reviewing the diff is the habit that matters.

Why is the diff different from `git diff`?

Because it is measured from the base commit, not from HEAD. The thread diff compares the working tree with the merge-base of your branch and its base branch, so it includes commits you have already made on the branch — what a reviewer would see. A plain git diff shows only what is uncommitted. Change base in the diff's bar is how you change what it measures against.

Are units converted anywhere?

No. Ferro inputs are strictly SI — Pa, N, m, K, kg, Hz — with no conversion layer. A case in millimetres solves without complaint and gives you a wrong answer.

Why is gmsh not included?

It is GPL-licensed, so it is not in the download. Settings → Local tools installs it on macOS, Windows and x86_64 Linux, and an install of your own is picked up from PATH — see Environment Setup. Qubis's own OCCT-backed mesher covers the common path without it.

FAQ