7. Interactive commands

You talk to LatticeMind in plain language. Every command below is optional — there is nothing here you cannot also ask for in a sentence. The commands exist for the few things a sentence cannot do, such as answering a pending permission request or leaving the program.

Tip

/help lists them inside the app, and /examples shows ready-to-run prompts.

7.1. In a research conversation

These are the commands of the default runtime. /help prints the short list:

Fig. 7.1.1 /help in a research conversation.

Command	What it does
/status	This session: its projects, the active turn, any pending permission.
/whereami	Project folder, provider, model, approval mode, solver configuration.
/events [n]	The last n recorded events, newest last.
/attach <path>	Introduce a file — a paper, a structure, a data file — as reference material.
/approve / /deny	Answer the pending permission request.
/cancel	Stop the active turn. A solver job already submitted keeps running.
/cancel-job <run>	Ask a submitted job to stop.
/approve-for-me on|off	Let a fixed set of rules answer routine requests (on), or answer every one yourself (off, the default). /approvals assessed and /approvals ask are the same switch.
/approvals-log	The approvals approve-for-me gave, with the rule behind each.
/model	Choose the model. Lists current models from the OpenAI, Claude, Gemini and Qwen APIs and a local Ollama, each labelled configured or unconfigured, and asks for a missing API key. /model <id> or /model <provider> <id> answers directly.
/remote	Choose where calculations run: saved profiles and hosts found in ~/.ssh/config, + to add one, l for this machine. /remote add, /remote test <profile>, /remote <profile> and /remote local answer directly; /remote setup and /remote warm prepare SSH session reuse for hosts with two-factor login.
/report [path]	Write an evidence report (HTML and Markdown) for the latest answer under results/, and print where.
/new	Start a fresh conversation in this folder.
/examples	Short, ready-to-run example prompts.
/help	The command list.
/exit	Leave LatticeMind.

7.2. Changing the model

/model shows a numbered list:

Current model: gpt-5.6-luna (OpenAI API)

API providers
  OpenAI API                  configured - OPENAI_API_KEY is set
     1  gpt-5.6-sol           (configured)          latest frontier model
     ...
  Claude API                  unconfigured - needs CLAUDE_API_KEY
     7  claude-fable-5-1      (unconfigured)        most capable
     ...
Local models
  Ollama (on this machine)    configured - running at http://127.0.0.1:11434
     ...

Fig. 7.2.1 /model in the terminal: every provider, whether it is configured, and the model in use. Enter keeps the current model.

Type a number or a model id. Choosing an unconfigured API model asks for its key (the input is hidden) and saves it to the .env file you are using: the project folder’s own .env when there is one, otherwise ~/.latticemind/.env. A key that a file already sets is updated in that file. The provider and the model are saved too, so the next conversation and the next start of LatticeMind use them; each provider remembers its own model (LATTICEMIND_<PROVIDER>_MODEL).

A conversation stays on the model it began with, so switching after the first message starts a fresh conversation in the same folder. The earlier one and its projects are kept. In the browser, click Model in the top bar, or type /model; the same list opens with a key field for an unconfigured provider.

7.3. Choosing where calculations run

/remote lists your saved remote profiles and the hosts in ~/.ssh/config, each labelled configured (a saved profile) or found (a host you could add):

Fig. 7.3.1 /remote in the terminal: saved profiles and SSH hosts, + to add one, l for this machine.

/remote                  list, choose, + add, l for this machine
/remote add              a short wizard: name, host, work folder, transfer
                         tool, solver commands, shell setup
/remote test office      check SSH, file transfer and the solver command;
                         nothing is uploaded
/remote office           send this conversation's calculations there
/remote local            run on this machine again
/remote setup office     add SSH session reuse (for two-factor login)
/remote warm office      open one reusable SSH session; approve 2FA once

The conversation and the model stay on this machine; only the solver runs on the remote host. Choosing a profile launches nothing, and each launch still asks for your permission unless approve-for-me covers it. To start with a profile already chosen, run latticemind --remote-profile office. Conversations launch over direct SSH; Slurm profiles are for latticemind quick. See Remote & HPC execution for the profile fields.

7.4. Answering a permission request

Before a solver runs, LatticeMind shows exactly what will run and where, and offers:

1. yes   2. yes, always   3. no

Type the number, or answer in words — yes, go ahead, run it, approve, or no, don’t, cancel. /approve and /deny are the same answers written as commands.

7.4.1. Approve-for-me

/approve-for-me on (or /approvals assessed) lets a fixed set of rules answer routine requests for the rest of the session:

Goes ahead without asking	Always asks you
A prepared solver run (RESCU, NanoDCAL, VASP) of low or medium estimated cost, on this machine or on a saved remote profile, with validation clean — one run or a batch.	A high-cost run on this machine, or one whose size cannot be estimated.
Creating folders or empty files inside the project (mkdir, New-Item -ItemType Directory, touch).	Moving, renaming, copying or deleting files and folders, writing into a file, and reorganising the project folder.
	Running a program or a generated analysis script, and anything outside the project folder.

Every approval it gives is recorded with the rule that allowed it; /approvals-log lists them. After twelve runs or twenty folder creations since your last message, the next request asks you, so a turn that keeps repeating itself cannot run on unattended; your next message resets the count. /approve-for-me off returns to answering everything yourself.

Fig. 7.4.1 /approve-for-me on in the terminal, with what goes ahead and what still asks.

In the browser, the Approvals fact in the top bar shows who answers; click it to switch. The Control room’s Permissions card has the same switch. To start a terminal session in approve-for-me, run latticemind --approvals assessed.

7.5. Steering the work

You do not need a command to change what LatticeMind is doing — say it, either at the prompt or while a turn is running:

Use a denser 8x8x8 k-grid for the SCF.
Add a DOS step after the SCF.
Relax the structure before the band structure.
Redo the band structure along L-Γ-X-W-K.
Prepare the inputs but do not run anything.

A message typed while a turn runs is read at its next step. ctrl-c offers s to stop, d to leave the work running, and w to keep watching.

7.6. The latticemind quick palette

The one-shot pipeline (Two runtimes) has a larger set of commands, listed inside it with /commands. They control a pipeline run rather than a conversation.

Command	What it does
/commands	Show the full pipeline command palette.
/new / /projects / /resume last	Start, list, or continue a pipeline project.
/attach . / /workspace	Adopt the current folder, or inspect it before deciding.
/trust [yes|no|forget]	Whether LatticeMind may use structure files it finds in the current directory. Files you attach by name are always used.
/title <name> / /fork last	Name a project; copy one before experimenting.
/execute on|off	Allow the solver to run after validation passes.
/launch-preview / /cost	Preview the exact commands and inputs; estimate the resource risk.
/compute <local|remote|slurm>	Choose where calculations run.
/rescu <command>	Point LatticeMind at your RESCU executable.
/lease status	Whether a released calculation still holds the project’s execution lease; /lease clear removes a stale one.
/provider / /model	Show or switch the AI model provider and model.
/status	Workflow and per-calculation status.
/inputs / /outputs / /plots / /analysis	The generated input decks; results, logs and reports; figures; extracted quantities.
/failed / /stale	Calculations that failed or became stale, with suggested reruns.
/continue <target>	Resume a failed or stale step.

7.7. Where to go deeper

• The complete pipeline palette, with every command and alias, is in The latticemind quick command palette.

• Remote/HPC commands (/remote …) are covered in Remote & HPC execution.

• Cloud/proxy commands (/qbraid, /cloud status) are in Cloud & headless deployment.