Semalt.AI Code CLI

@semalt-ai/code is a self-hosted AI coding assistant for the terminal.

It provides an interactive chat interface, one-shot code generation, AI-assisted file editing, shell command execution, and an agent loop that can read files, write files, and run commands after user approval.

Features

• Interactive terminal chat mode
• OpenAI-compatible API integration
• Streaming responses with terminal-friendly formatting
• Optional reasoning stream display when the backend provides it
• User-approved shell, file read, and file write actions
• File and directory context loading
• One-shot code task mode
• AI-assisted file editing mode
• Model discovery from the configured API

Requirements

• Node.js >=16
• An OpenAI-compatible API endpoint

The default configuration expects a local API server at http://127.0.0.1:8800.

The CLI accepts api_base in either of these forms:

• http://127.0.0.1:8800
• http://127.0.0.1:8800/v1

Both formats are normalized automatically.

Installation

Install the package globally so the semalt-code command is available system-wide.

npm install -g @semalt-ai/code

After global installation, run:

semalt-code

Initial Setup

Create the CLI config:

semalt-code init --api-base http://127.0.0.1:8800 --api-key any --dashboard-url https://cli.semalt.ai --default-model default

This writes configuration to:

~/.semalt-ai/config.json

Example config:

{
  "api_base": "http://127.0.0.1:8800",
  "api_key": "any",
  "dashboard_url": "https://cli.semalt.ai",
  "auth_token": "",
  "default_model": "default",
  "temperature": 0.7,
  "request_timeout_ms": 900000,
  "stream": true
}

You can also set "api_base" to a URL that already ends with /v1.

Usage

semalt-code [command] [options]

Commands

• semalt-code

• semalt-code chat Starts interactive chat mode.

• semalt-code code <prompt> Runs a one-shot coding task.

• semalt-code edit <file> <instruction> Sends a file to the model and overwrites it with the returned result.

• semalt-code shell <command> Runs a shell command with approval prompts.

• semalt-code login Starts browser-based login and stores the confirmed CLI token in ~/.semalt-ai/config.json.

• semalt-code whoami Shows the current user associated with the saved CLI auth token.

• semalt-code logout Logs out the current CLI user and clears the saved local auth token.

• semalt-code models Lists your models from the dashboard and lets you choose the current one for the CLI.

• semalt-code models add Opens an interactive flow to add an API base URL, API key, and model ID as a reusable model profile.

• semalt-code init Creates or updates the local config file.

Options

• -m, --model <name> Override the model name.

• -v, --version Print the current CLI version.

• -f, --file <path> Load one or more files or directories into the prompt context for code.

• -a, --analyze After shell, ask the model to summarize the command output.

• --dry-run For edit, show the generated result without saving it.

• --api-base <url> Set the API base URL during init.

• --api-key <key> Set the API key during init.

• --default-model <name> Set the default model during init.

Interactive Chat Mode

Running semalt-code without arguments starts the terminal chat UI.

Available interactive commands:

• /help
• /file <path>
• /whoami
• /logout
• /model
• /login
• /model <name>
• /models
• /clear
• /compact
• /cost
• /shell <cmd>
• !<cmd>
• /approve
• /config
• exit

What /file does

• If you pass a file, its full content is added to the conversation context.
• If you pass a directory, the CLI recursively loads up to 50 non-hidden files.

Agent Behavior

The assistant is instructed to use special tool-like tags:

• <exec>...</exec> for shell commands
• <read_file>...</read_file> for file reads
• <write_file path="...">...</write_file> for file writes

When the model emits these actions, the CLI:

1. Detects them in the response
2. Prompts the user for approval
3. Executes the action
4. Sends the result back to the model
5. Continues for up to 10 iterations

This makes the tool behave like a lightweight terminal agent while keeping the user in control.

Examples

Start chat

semalt-code

Ask for a coding task with file context

semalt-code code -f package.json -f index.js "Explain this project and suggest improvements"

Edit a file with AI

semalt-code edit index.js "Refactor duplicated logic into helper functions"

Preview an edit without saving

semalt-code edit index.js "Add better error handling" --dry-run

Run and analyze a shell command

semalt-code shell -a "npm test"

List models

semalt-code models

Add a saved model profile

semalt-code models add

The CLI will ask for:

• API Base URL
• API Key
• Model ID

Each saved profile is appended to the profile list.

Saved profiles can then be selected inside chat mode with /model or /models.

Show the current version

semalt-code --version

How Responses Are Rendered

The CLI formats streamed output for terminal readability:

• headings
• bullet lists
• numbered lists
• fenced code blocks
• diff-like output
• inline code and file paths

If the backend returns reasoning_content, the CLI also shows a lightweight thinking section during streaming.

Notes and Limitations

• This project is currently a single-file CLI implementation centered in index.js.
• It uses Node's built-in http and https modules and does not require extra runtime dependencies.
• The edit command writes the model output directly back to the target file, so review prompts and backend behavior carefully.
• Shell and file operations are approval-based, but they still execute on the local system after approval.

License

MIT