2. Terminal and browser

A research conversation has two interfaces: the terminal and the browser. Both show the same conversation, so you can start in one and continue in the other. The one-shot pipeline is separate and is started with latticemind quick (see Two runtimes).

Interface

Best for

How to start

Terminal

Laptops, workstations, HPC login nodes and scripts

latticemind in the project folder

Browser

A visual view of the conversation, files, runs and figures

latticemind-web (see The web console)

Note

The earlier terminal dashboard has been retired. The earlier browser dashboard built for the one-shot pipeline is still served at /classic.

2.1. Starting and reopening a conversation

cd silicon-bands
latticemind                          # terminal
latticemind-web --port 7865 --no-open    # browser

To reopen a conversation, start latticemind in the same folder again, or choose the folder in the browser and pick the conversation from the list it offers. A conversation keeps the model and the execution settings it began with; /new starts a fresh conversation in the same folder, and switching models with /model does the same. Earlier conversations and their projects are kept.

Figures appear inline in the browser and, where the terminal supports images, in the terminal. They are always saved under results/plots/.

2.2. Which should I use?

  • On your laptop: either. The browser shows files, runs and figures side by side; the terminal is quicker to start.

  • On an HPC login node: the terminal. Calculations can go to a remote profile (/remote) without a browser on the compute host.

  • From a shell agent or a script: see Using LatticeMind from Codex, Claude, and shell agents. JSON output gives machine-readable results; it does not grant permission to run a solver.

The command palette of latticemind quick is documented in The latticemind quick palette. Its controls are not the same as a conversation’s.