LS Code for Windows, macOS and Linux

The coding agent that finishes the job

LS Code lives in your terminal. It reads your project, edits files, runs your commands and tests, and keeps going until the work is checked and done, on models that do what you ask.

curl -fsSL https://lessstrict.com/install | sh

Paste it into your terminal. Then run lscode login once, and lscode in any project folder.

Latest version: 0.27.1. Already installed? lscode --version shows yours, and lscode update brings it up to date.

  • One command to install
  • Nothing else to set up
  • Asks before it changes anything

◆ LS Code

~/projects/shop-api

Built for work you can walk away from

Most coding assistants stop when they think they are done. LS Code stops when the work has been checked.

Goal mode with a judge

Give it a goal with /goal. It writes down what finished means, works through it, and a separate judge model checks every point before it is allowed to say done.

Workers that run side by side

Large jobs are split between workers that each get their own copy of the project, so they never step on each other. Their changes come back to you to keep or drop.

Control it from anywhere

Turn on /remote and watch the session from your dashboard on any device. Send the next task with files attached, answer its questions, or stop it.

Online in one command

/deploy puts a static site on its own lessstrict.app address in seconds. Run it again to update, and take it down whenever you like.

Your secrets stay yours

Server and database logins live in your system keychain. The AI works on them by nickname and is never shown an address or a password. A firewall hides keys it comes across in files.

It does not give up

When a connection drops or a provider has a bad minute, LS Code waits and tries again, then carries on from where it was. No more typing continue.

Install LS Code in a minute

One program, no dependencies. It checks what it downloads and keeps itself up to date.

  1. 1

    Run the install command

    It downloads the right build for your system and puts lscode on your path.

    curl -fsSL https://lessstrict.com/install | sh

    Paste it into your terminal. Then run lscode login once, and lscode in any project folder.

    Latest version: 0.27.1. Already installed? lscode --version shows yours, and lscode update brings it up to date.

  2. 2

    Sign in once

    Run lscode login and approve it in your browser. The computer stays signed in until you remove it.

    lscode login
  3. 3

    Start it in a project

    Open a project folder, run lscode, and describe what you want done.

    cd my-project
    lscode

Everything it does

The parts you will use every day, and the ones you will be glad are there.

A real agent, in the terminal you already use

  • Reads and searches your project, edits files with a diff you can see, and runs your commands and tests.
  • Four modes, switched with Shift+Tab: ask before each change, edit freely, full access, or plan only.
  • Long commands run in the background while it keeps working, and it can ask you a question with choices when it needs a decision.
  • Paste a screenshot with Alt+V or name an image file, and models that can see will look at it.

The right model for each step

  • Pick any model on your LessStrict account with /model, and its thinking effort with /effort.
  • Auto routing sends simple steps to a faster model and keeps the strong one for the hard parts.
  • /race gives the same task to several models at once, each in its own copy of the project, so you can keep the best result.
  • /budget caps what a single task may cost, and /cost shows what the session has used.

Works on your servers and databases, safely

  • Save a server once and the AI can run commands on it over SSH by its nickname.
  • Save a PostgreSQL or MySQL database and it can query it. Read-only by default; every change asks you first.
  • Addresses, user names and passwords never leave your computer and are never put in front of the model.
  • Send traffic through your own proxy always, or only when a site blocks a direct connection. Nothing on your computer changes.

Sessions that survive

  • /resume picks up any earlier session in the folder, with its whole conversation.
  • /rewind undoes the last turns, files included. /compact shortens a long conversation so there is room to keep going.
  • /detach leaves a session running after you close the window. Come back to it later, or from the dashboard.
  • /schedule starts work by itself: nightly test runs, a morning summary, a weekly dependency check.

Skills and tools included

  • Built-in skills for planning, debugging, test-first work, reviews, security, front-end design, refactoring, commits and more. All stay off until you switch them on, so they cost nothing until needed.
  • Add your own skill as a folder with a SKILL.md file, and run any skill by name.
  • MCP servers plug in more tools. Library documentation is ready on first run; a browser, GitHub, Semgrep and others are one step away.
  • Hooks run your own commands after edits and when a task ends, such as a formatter or a linter.

Remote control you can lock

  • Set a passphrase with /remote lock and everything sent to the dashboard is encrypted on your computer first.
  • The passphrase is typed in your browser and never sent to us. Without it, nobody can read or steer the session, not even with your account.
  • Only a narrow set of actions is accepted from the web: send a task, answer a question, approve a step, stop.
  • Connect a Telegram bot of your own to talk to it from your phone, or get a Slack or Discord message when a task finishes.

Skills for the way you work

Packaged know-how the agent loads when a task matches, from test-first coding to security. Every skill is off until you switch it on in /settings, so nothing slows you down until you want it.

Plan and build

Think before changing code, and change it in safe steps.

/plan
Pin down what is wanted and write a short plan before building. On by default.
/debug
Find the real cause of a bug before touching the code.
/tdd
Write the failing test first, then the code that passes it.
/refactor
Restructure code without changing what it does, in verified steps.
/verify
Run the checks and report exactly what was and was not confirmed.

Review and security

Check code and systems for weaknesses, on the offensive and defensive side.

/code-review
Review a diff or branch for real bugs and risks.
/security-review
Trace untrusted input to find injection, access and secret flaws.
/pentest
Test a site or API you point it at: recon, probe, report.
/dependency-audit
Find known-vulnerable packages and propose safe upgrades.
/secrets-scan
Find keys and passwords in the code or git history, and rotate them.
/hardening
Review a server's setup: SSH, firewall, updates, ports, permissions.
/threat-model
Work out what could be attacked, and what to do about each risk.

Ship and document

Get the work out and keep it understandable.

/git-commit
Stage the right changes and write a clear commit or PR description.
/write-docs
Write a README, guide or API reference a newcomer can use.
/deploy-site
Build a web project and put it online at a lessstrict.app address.
/performance
Measure first, fix the biggest cost, measure again.

Data, servers and design

Work on the things around the code, carefully.

/database
Explore a saved database and change data or structure safely.
/server-ops
Work on a saved server: inspect first, change with a way back.
/frontend-design
Design websites, apps, dashboards, forms and emails to exact rules, then check the real page at phone and desktop width. On from the start.
/skill-creator
Make a new skill of your own for LS Code to load when it fits.

Plus your own: drop a SKILL.md in your project or home folder and it appears alongside these. A skill you add with a built-in's name takes its place.

Auto routing

A faster model takes the simple steps.

Background jobs

Dev servers and long builds keep running while it works.

Memory

It remembers what you tell it to, per project and for you.

Git aware

Workers use separate working copies, so your branch stays clean.

Live refresh

/refresh picks up new models and settings without restarting.

Private errors

You see a clear message; provider details stay out of sight.

Mouse and keyboard

Scroll, select and copy text, and view the model's thinking.

Cost in view

Tokens and dollars are on the status line the whole time.

Everything is a slash away

Type / in LS Code to see every command. These are the ones people reach for first.

/goal
Work until a judge confirms the goal is met
/model /effort /mode
Choose the model, how hard it thinks, and how much it may do without asking
/race
Run a task on several models and keep the best
/remote /detach
Watch and control the session from your dashboard
/deploy /sites
Publish a folder as a site, and manage your sites
/server /db
Save servers and databases the AI can work on by nickname
/resume /rewind /compact
Go back to a session, undo turns, make room
/skills /mcp /hooks
Built-in skills, extra tools and your own commands
/schedule /jobs
Work that starts by itself, and commands running in the background
/settings
Everything you can switch on, off or change, in one list

Questions about LS Code

What is LS Code?

LS Code is an AI coding agent that runs in your terminal. You describe a task in plain words and it reads your project, edits files, runs commands and tests, and reports back. It uses the models on your LessStrict account.

How do I install LS Code?

Paste one line into your terminal. On macOS and Linux: curl -fsSL https://lessstrict.com/install | sh. On Windows, in PowerShell: irm https://lessstrict.com/install | iex. Then run lscode login once to connect your account, and lscode in any project folder.

Which systems does it run on?

Windows, macOS and Linux, on both Intel/AMD (x64) and ARM chips, including Apple silicon. It is a single program with nothing else to install, and it updates itself with lscode update.

What does it cost?

The app itself costs nothing. The AI usage counts against your LessStrict plan, or your wallet at each model's per-token price, exactly like the chat and the API. /cost shows what a session has used and /budget sets a limit per task.

Is my code safe?

LS Code runs on your computer and only sends the model what a task needs. It asks before it changes files or runs commands, in the default ask mode; switch to auto-edit to let it change files in the project freely, to full access to let it do anything, or to plan mode to only read. Server and database logins stay in your system keychain and are never shown to the model, and a built-in firewall hides keys and passwords it finds in files.

Can I use it from my phone?

Yes. Type /remote on in a session and it appears on the Sessions page of your dashboard, on any device. You can watch it work, send the next task with files, answer its questions and stop it. A passphrase lock encrypts everything end to end.

What makes it different from other coding agents?

It works with less restrictive open-source models, it will not call a task finished until a separate judge has checked it in goal mode, it can split work between parallel workers or race several models, it publishes sites in one command, and it keeps your server and database credentials away from the AI.

Does it work with MCP servers?

Yes. LS Code connects to MCP servers over a local program or a web address. A catalog of well-known ones is added in one step with /mcp, including GitHub for issues and pull requests and Semgrep for security scanning, and you can add your own. It tells you when they connect.

What skills come with it, and can it help with security?

It ships with skills for planning, debugging, test-first work, code review, refactoring, docs, commits, deploys, databases and servers, plus a security set: security-review, pentest, dependency-audit, secrets-scan, hardening and threat-model. Every skill is off until you switch it on in /settings, so you turn on only what you need. The security skills read code and systems for weaknesses and test them from the outside: /pentest takes the target you name, starts on it without a permission form, probes carefully, changes nothing, and reports each finding with the request that proves it.

Give it your next task

Install LS Code, open a project, and say what you want done.

curl -fsSL https://lessstrict.com/install | sh

Paste it into your terminal. Then run lscode login once, and lscode in any project folder.

Latest version: 0.27.1. Already installed? lscode --version shows yours, and lscode update brings it up to date.