2. Command-line & interactive commands

This page describes LatticeMind’s command-line entry points and the command palette of latticemind quick. A research conversation and the one-shot pipeline have different commands; for a task-first introduction to a conversation’s commands see Interactive commands.

2.1. Console commands

The wheel installs these commands on your PATH:

Command	Purpose
latticemind	Start a research conversation in the current folder (the default runtime). Running it again in the same folder reopens the conversation.
latticemind setup	Ask which model to use and where RESCU is, and save the answers to ~/.latticemind/.env.
latticemind setup --show	Print which providers and solvers are ready, and the settings file in use.
latticemind quick	Start the one-shot pipeline.
latticemind-web	Open the web console in your browser.

2.2. Starting a conversation

latticemind accepts a few options for the conversation it starts:

Option	Meaning
--approvals assessed	Start with approve-for-me on (ask, the default, asks about every launch).
--remote-profile <name>	Send this conversation’s calculations to a saved remote profile. The same as /remote <name> after starting.

LATTICEMIND_RUNTIME=quick makes latticemind start the one-shot pipeline instead. The retired terminal dashboard is not available; latticemind-web is the graphical interface.

For scripts, latticemind research drives a conversation one message at a time (see Drive a research conversation from a script). latticemind research chat --help lists its options. For example, local VASP through an existing WSL installation:

latticemind research chat --local-solver vasp --solver-executable /usr/local/bin/vasp_std --vasp-wsl-distribution Ubuntu-24.04 --solver-threads 1 --solver-timeout 300 --allow-project-changes

Replace the distribution and executable with your installed values, and set VASP_PP_PATH or LATTICEMIND_VASP_PP_PATH to your licensed POTCAR library. Local RESCU and NanoDCAL use the same --local-solver and --solver-executable options without the WSL option; a remote host uses --remote-profile NAME --remote-solver rescu|nanodcal|vasp instead. Choosing an executable does not approve a launch: each one still asks.

2.3. Shell discovery and one-shot commands

The latticemind command also provides non-interactive commands for shell users and general agents. Add --json or --format json anywhere in a command to receive the versioned JSON response documented in Using LatticeMind from Codex, Claude, and shell agents. latticemind run is the one-shot interface and works on latticemind quick projects.

Command	Purpose
latticemind --version	Print the installed LatticeMind version without a license/provider check.
latticemind capabilities --json	Return solvers, shell commands, slash commands, exit codes, and execution safeguards.
latticemind commands [filter] --json	Return the full interactive command catalog or one category/search match.
latticemind docs <question>	Print one concise answer after inferring LatticeMind, RESCU, NanoDCAL, or all corpora from the question.
latticemind docs --list	List packaged LatticeMind manual source paths.
latticemind docs --source <path>	Read one exact packaged LatticeMind manual .rst page.
latticemind protocol --json	Return the versioned protocol manifest.
latticemind doctor --json	Run read-only local and remote-configuration checks.
latticemind project inspect [path] --json	Read a latticemind quick project’s workflow, calculation status, and indexed files.
latticemind project ask <question> --project <path> --json	Answer from existing project files and matching LatticeMind manual sections without writing to the project.
latticemind run <prompt> --json	Run one one-shot planning and validation turn; the solver does not run by default.
latticemind run <prompt> --execute --approve-execution --json	Run one turn with explicit permission for the validated solver launch.
latticemind run <prompt> --execute --approve-execution --detach	Release the approved calculations and return immediately. See Releasing a long calculation (latticemind quick).
latticemind run <prompt> --no-subdir	Work in the current folder instead of creating a project subdirectory. Home, root, system, and software folders still get a subdirectory.
latticemind run <prompt> --solver {rescu,nanodcal,vasp}	Bind a new project to one solver. Left unset, a new project uses RESCU and an existing project keeps the solver it was created with.
latticemind run <prompt> --remote-profile <name>	Execute on the named profile from ~/.latticemind/remote_profiles.toml instead of this machine. See Remote & HPC execution.
latticemind run <prompt> --project <dir> / --new-project	Continue an existing project, or create a new one even when the current directory could be attached. A continuation inherits the device, material, and settings of the request the project was founded on.

See Using LatticeMind from Codex, Claude, and shell agents for the recommended Codex/Claude command sequence, JSON fields, exit codes, project continuation, and license/provider setup.

Solver documentation searches return excerpts only; --source reads only this manual, not an exact RESCU or NanoDCAL page. docs uses --corpus auto by default. Use --corpus <name> to override the inferred corpus, --json for a compact machine-readable response, and --verbose for full fields on the returned excerpts. --limit N changes the maximum number of returned excerpts.

2.4. Releasing a long calculation (latticemind quick)

In a research conversation a long run needs nothing special: close the terminal and the solver keeps running, and latticemind in the same folder picks the results up. The one-shot interface has an explicit option instead. Adding --detach releases the approved calculations to run independently — as a Slurm job, an SSH job, or a local background process — and returns immediately:

latticemind run "scf + bs + dos for Si: mp-149" \
    --execute --approve-execution --detach

You can then close the terminal, log out, or shut down the client machine if the work is running remotely. Pick the project up later with:

latticemind run --project <project-dir> "continue"

LatticeMind reattaches to the released work, collects whatever finished while it was away, and carries on from there. Inside latticemind quick, /status reports what is running and what has completed, and /stale reports steps whose inputs changed and therefore need recomputing. If a launch says it is waiting for released work you know has finished, /lease status shows what is recorded and /lease clear lets the next run proceed.

This is the recommended way to run anything long in the one-shot interface — geometry relaxations, phonon dispersions, or large supercells — and the only practical way to use a batch scheduler, where the job may not start until long after you submit it.

2.5. latticemind-web options

Flag	Meaning
--host <addr>	Bind host. Default: LATTICEMIND_WEB_HOST or 127.0.0.1; cloud mode keeps 127.0.0.1; notebook platforms proxy to it.
--port <n>	Bind port. Default: LATTICEMIND_WEB_PORT or 7865.
--no-open	Do not attempt to open a local browser (use on servers/cloud).
--open-browser	Force opening a local browser.

Example:

latticemind-web --port 7865 --no-open

2.6. Remote utility subcommands

Before opening the interactive app, you can manage remote SSH sessions directly from the shell. These are handled by latticemind at startup:

latticemind remote setup <profile>    # install SSH session-reuse settings
latticemind remote warm <profile>     # open one approved SSH session (2FA once)
latticemind remote status <profile>   # check recorded remote job status
latticemind remote close <profile>    # close the saved SSH session early

See Remote & HPC execution for the full remote/HPC workflow.

2.7. The latticemind quick command palette

In a research conversation, /help prints the short list of the conversation’s own commands (see Interactive commands), and /commands opens this larger palette. The palette was written for latticemind quick. Most of its commands also answer in a conversation, and where a command means something different there, it says so. The tables below describe the pipeline meaning, with the conversation’s meaning noted where it differs.

Help

Command	Description	Example
/commands	Show the command palette. In latticemind quick, /help is an alias; in a conversation, /help is the short list.	/commands
/help <topic>	Explain a command or topic.	/help execution
/why	Explain the current pause, blocker, or suggested action (alias ?).	/why
/tips	Show context-aware next actions.	/tips
/examples	Show short example prompts.	/examples

Project

Command	Description	Example
/new	Start a fresh project (quick). In a conversation: a fresh conversation in this folder.	/new
/projects [limit]	List recent indexed projects.	/projects 8
/resume <project>	Resume a project by id, prefix, or path.	/resume last
/fork <project>	Fork a project before experimenting.	/fork last
/attach .	Adopt the current folder as the active project.	/attach .
/subdir [on|off]	Quick only. Create a project subdirectory for new work (on, the default) or work in the current folder (off). A conversation always works in the folder it was started in.	/subdir off
/attach-folder <path>	Adopt an existing folder without copying it.	/attach-folder .
/attach <file...>	Attach documents/files to the active or next project.	/attach notes.pdf
/workspace	Inspect the current folder before choosing new/resume/attach.	/workspace
/title <name>	Set a readable display title (alias /rename).	/title Silicon DFPT
/whereami	Show active project, mode, backend, settings (alias /project).	/whereami
/status	Show workflow and calculation status.	/status

Files

Command	Description	Example
/tree	Show key project folders and artifacts.	/tree
/inputs	List known solver input files.	/inputs
/outputs	List known results, plots, logs, and reports.	/outputs

Workflow & validation

Command	Description	Example
/failed	List failed calculation instances and suggested reruns.	/failed
/stale	List stale calculation instances.	/stale
/continue <target>	Continue a failed/stale workflow target.	/continue dfpt_phonon_bs
/rerun <target>	Force a target to rerun.	/rerun band_structure
/validator <on|off>	Quick only. Enable or skip validation. A conversation always checks its inputs.	/validator on
/readiness	Show launch probability and gate findings.	/readiness
/contract	Show workflow-contract summary.	/contract
/warnings	Show validation/readiness warnings.	/warnings

Execution

Command	Description	Example
/execute <on|off>	Quick only. Enable or disable solver execution. A conversation asks before each launch instead.	/execute on
/approve-for-me <on|off>	Let routine, low-risk work go ahead; anything else still asks. In a conversation, see Interactive commands for exactly what goes ahead.	/approve-for-me on
/rescu <command>	Set the RESCU executable or launch command.	/rescu /opt/esapp/rescu
/compute <local|remote|slurm>	Choose local, SSH, or Slurm execution (quick). In a conversation, use /remote; Slurm is not available there.	/compute slurm
/launch-preview	Preview what would run before launch (alias /dry-run).	/launch-preview
/cost	Estimate resource risk and staged-run recommendations.	/cost
/rescu-plotting [on|off]	Enable or disable RESCU’s built-in plotter (rescu -p); the choice is saved with your settings. Disable it on machines without a MATLAB Java runtime (JVM) — Python post-processing still generates every figure.	/rescu-plotting off
/lease status	Quick only. Show whether released work still blocks the next launch (alias /lease). See Releasing a long calculation (latticemind quick).	/lease status
/lease clear	Quick only. Let the next run launch after released work you know has finished. Refuses while that work is still running unless given force.	/lease clear

Remote / HPC

Command	Description	Example
/remote wizard	Create or update a remote profile with guided prompts (in a conversation: /remote add).	/remote wizard
/remote profiles	List remote profiles.	/remote profiles
/remote profile <name>	Select a remote profile.	/remote profile cluster
/remote setup [profile]	Install SSH session-reuse settings.	/remote setup cluster
/remote warm [profile]	Open a reusable SSH session for 2FA.	/remote warm cluster
/remote test [profile]	Run SSH/transfer/scheduler preflight.	/remote test cluster
/remote doctor	Explain remote readiness and next checks.	/remote doctor
/remote status	Check recorded remote job status.	/remote status
/remote fetch	Fetch recorded remote job outputs.	/remote fetch
/remote logs [lines]	Tail latest remote execution logs.	/remote logs 80
/remote workdir	Show remote work directory and local bundle paths.	/remote workdir
/remote cancel	Cancel the recorded remote job.	/remote cancel

Post-processing

Command	Description	Example
/postprocess	Run or preview post-processing scripts.	/postprocess
/plots	List generated plots.	/plots
/analysis	Show the analysis summary.	/analysis

Runtime, cloud & session

Command	Description	Example
/provider [name]	Show the active AI model provider and models, or switch to another (alias /llm). In a conversation, use /model.	/provider
/model [id|off]	Quick: show the active model, pin an exact model id, or clear the pin. In a conversation: choose from every provider’s models (see Interactive commands).	/model
/doctor	Read-only environment diagnostics (CLI, web, RESCU, remote).	/doctor
/runtime	Inspect or persist runtime/deployment context.	/runtime
/language <auto|en|fr|zh>	Choose the final report language (alias /report-language).	/language auto
/qbraid	Show the qBraid Web UI and proxy quickstart.	/qbraid
/cloud status	Show cloud/headless web settings.	/cloud status
/web	Launch the web console.	/web
/exit	Exit LatticeMind.	/exit

2.8. Reports

In a conversation, /report writes an evidence report for the latest answer — HTML and Markdown — under results/ and prints where. The same report can be written from the shell:

latticemind research report --session SESSION_ID
latticemind research report --session SESSION_ID --kind answer
latticemind research report --session SESSION_ID --markdown custom.md --title "Custom research summary"

--kind evidence is the default. --kind answer includes the selected answer’s text. --turn TURN_ID selects an older answer. Custom Markdown may quote recorded values with the conversation’s {{fact:...}} references, which are checked against the recorded results; the methods, results, limitations and sources are always appended, and prose you write yourself is not independently checked. Open the HTML locally, or use the browser’s Print / Save as PDF. Local figures under the project are embedded; scripts, remote images and network fonts are not loaded. An existing report is never overwritten.

Automatic reports in latticemind quick. The one-shot pipeline writes a report after each run by default. --report-format off skips it but keeps chat and calculation results; previous reports are not deleted. Choices can be saved per project:

latticemind report configure --project my_project --automatic off
latticemind report configure --project my_project --automatic on --title "Silicon study" --font sans --density compact
latticemind report configure --project my_project --brief "For a collaborator: findings first, concise methods, explicit limitations."

--automatic default restores the default, and --accent '#125678' changes the decorative accent colour. A brief guides future model-written pipeline reports and does not hide integrity findings. To render your own project-relative Markdown without a model or solver call:

latticemind report create --project my_project --markdown collaborator.md --title "Collaborator summary"

2.9. Environment variables

For runtime environment variables, see Settings & configuration reference for the complete list.