Skip to content

FAQ

Yes, while it is in private preview. There is no account.

LAMMPS today. Simulations, versions, runs, environments, executables, checks and the agent are not tied to one engine; what is engine-specific is the check rewriting, the build spec, the executable facts and the documentation index. More engines are planned.

No. On remote machines everything goes into ~/.verlet. The only optional exception is profiling, which needs one setcap on the verlet-probe binary.

Your simulations and history stay in verlet-agent’s database on your computer. Inputs go to the machines you run them on. The agent sends your description, retrieved documentation and its drafts to the model API you configure. Use a local model server if that must not leave your computer. No telemetry, no account, no verlet server.

Any OpenAI-compatible API: OpenRouter (the default), hosted providers’ compatible endpoints, or a local server such as Ollama or LM Studio. Larger models write better scripts; every script is checked with the engine either way.

Today, on the login node over SSH. Slurm submission is planned. Check your centre’s rules for login nodes.

My machine has no engine. Do I need to build it myself?

Section titled “My machine has no engine. Do I need to build it myself?”

No. Set up the environment and create a build from a template. verlet compiles it in your home with Spack.

Only when you save in its editor, and only after a passing check. Runs and checks use copies: the run copy flushes thermo output on every line, the check copy takes no timesteps.

They are read from where they are. Only the input script is versioned and uploaded, so files a remote run needs must already be on that machine at the path the script uses.

Yes: the demo is the real app in your browser with sample data. Nothing runs and nothing is saved.