LetCode is an AI coding agent: it understands your project, edits code, runs commands and passes tests — from the desktop, the terminal or the browser. Built on a deeply re-engineered agent kernel, with models, tools and capabilities composed entirely through plugins. All mainstream AI models are supported.
This manual covers: installation & startup, interface tour, core features, model configuration, command reference, security and privacy. Written for newcomers who have never used an AI coding agent — and for developers who want to run LetCode inside team pipelines.
Understands project structure, coordinates cross-file edits, watches build and lint errors, and proactively fixes type mismatches and missing references.
Executes shell commands in your project and watches output live: install deps, run tests, build, deploy.
Plan mode explores, asks questions and proposes a strategy — you approve before it executes. Every change requires your sign-off.
Scheduled tasks, headless pipelines, multi-agent teams and messaging platforms — hand the repetitive work to the agent.
Download the source package, then run the one-click script in your terminal. It handles environment checks, dependency install, build and install automatically:
The script automatically: checks and installs Xcode Command Line Tools → Rust → Bun 1.4.2 → installs dependencies → builds → packages with Tauri → installs into Applications. First run takes about 10-20 minutes; after that, double-click to launch.
Artifacts live in src-tauri/target/release/bundle/ — drag the .dmg into Applications.
xattr -cr /Applications/LetCode.app to clear the quarantine flag (see FAQ).fix the token-expiry bug on the login pageThe LetCode desktop app puts sessions, agent activity, files, Git changes, terminal and plans into a single window:
| Area | What it does |
|---|---|
| Project sessions | Sessions are saved per project — switch repos and resume anytime; create new sessions, open settings |
| Chat | Type tasks in natural language; the agent replies, asks and proposes plans here |
| Terminal / activity | Live view of commands and results: test output, build logs, tool calls |
| Diff preview | Every file change shown as a diff — approve, edit or revert (checkpoint) |
| Files / Git | Browse workspace files and Git status to see what the agent is changing |
Plan mode only explores: it reads code, asks clarifying questions and proposes a strategy. Once you align, switch to Act mode and it starts editing files and running commands.
LetCode first reads your project structure to understand how files relate, then coordinates cross-file edits. As it works it keeps watching lint and build errors, proactively fixing missing references, type mismatches and syntax issues. Every change has a checkpoint — undo the agent's work at any time.
LetCode executes commands in your project terminal and watches output live: install deps, build, test, deploy, touch the database. For long-running processes like a dev server, it keeps working in the background and reacts to new output — compile errors, failing tests and crashes are caught immediately.
Create a .letcoderules file in your project to define rules: coding standards, architecture conventions, deployment flows, testing requirements. CLI, desktop and editor extensions all follow these rules automatically. Skills load finer-grained behavior on demand, so the model follows your process for specific tasks.
A coordinator splits complex tasks into subtasks and dispatches them to specialized agents — each with its own tools and context. Team state persists across sessions, so you can resume after interruptions:
Run recurring automation on a cron schedule: daily PR summaries, weekly dependency checks, repository health reports. Tasks survive restarts and don't depend on any terminal session:
Connect the agent to Telegram, Slack, Discord, WhatsApp, Google Chat or Linear: each thread maps to a full-context agent session, with access control:
Runs with zero interaction — perfect for scripts and pipelines: pipe input, JSON output, command chaining, straight into CI/CD:
LetCode isn't locked to any model vendor. Add the matching API key in Settings (or environment variables) to switch — you can change models per project, anytime.
| Provider | Available models |
|---|---|
| Anthropic | Claude Opus / Sonnet / Haiku |
| OpenAI | GPT series |
| Gemini series | |
| DeepSeek | DeepSeek series |
| Qwen / Kimi / GLM | Qwen, Kimi, Zhipu GLM and other leading models |
| OpenRouter | 200+ models behind one unified entry point |
| AWS Bedrock / Azure / GCP Vertex | All models hosted on the major clouds |
| Cerebras / Groq | High-speed inference models |
| Ollama / LM Studio | Local models (no key, data never leaves your machine) |
| Any OpenAI-compatible endpoint | Private or third-party APIs |
| Command | Description |
|---|---|
letcode "task description" | Run a task interactively in the terminal |
letcode --json "task" | Output structured JSON for scripts |
letcode --team-name <team> "task" | Run in multi-agent team mode |
letcode schedule create | Create a scheduled task (cron) |
letcode mcp | Manage MCP server connections |
letcode connect <platform> | Connect messaging platforms (Telegram / Slack etc.) |
git diff | letcode "review" | Pipe input — review changes |
| Command | Description |
|---|---|
@file path | Reference a project file in chat for precise context |
@symbol name | Reference a code symbol — jump to definitions & call sites |
/skills | Load skill packs that follow your workflows |
/mcp | Connect external tools: databases, browsers, internal systems |
/agents | Orchestrate parallel sub-agent tasks |
/share | Share a session with your team for seamless handoff |
xattr -cr /Applications/LetCode.app to clear the quarantine flag. Official distribution needs an Apple Developer signature and notarization.grpc-tools postinstall may occasionally fail on the network — just re-run bun install.For developers who want to re-engineer or contribute:
| Component | Tech |
|---|---|
| Desktop shell | Tauri 2 (Rust) |
| Frontend | Next.js (webview) |
| Runtime | Bun sidecar (self-contained letcode-cli binary) |
| Core SDK | Multi-package: shared / llms / core / agents / ui / sdk |
| Theme | Deep-ocean palette; edit sdk/packages/ui/theme/palette.css + tokens.css, then run generate:theme |