Quickstart

Build Claurst from source and run your first AI coding session in under five minutes.

You can configure Claude’s behavior for a specific project by adding a CLAUDE.md file to your project root or to ~/.claude/CLAUDE.md for global settings. Claurst automatically discovers and injects these files into the system prompt at startup. See CLAUDE.md configuration for details.

1. Build from source

   Clone the repository and build the release binary with Cargo.

   git clone https://github.com/yocxy2/claurst
   cd claurst/src-rust
   cargo build --release
   

   This produces a single binary at target/release/claude. Building in release mode is recommended — it enables full optimizations and produces a significantly faster binary.

   For full installation options including PATH setup and pre-built binaries, see the Installation guide.

2. Set your API key

   Claurst connects to the Anthropic API. Export your API key before running the binary.

   bash / zsh

   export ANTHROPIC_API_KEY="sk-ant-..."
   

   To make this permanent, add the export line to your ~/.bashrc, ~/.zshrc, or equivalent shell profile. You can also pass the key inline at runtime with --api-key.

   Get an API key from the Anthropic console.

3. Open the interactive terminal

   Run claude from any project directory to open the interactive TUI.

   ./target/release/claude
   

   Or, if you have added the binary to your PATH:

   claude
   

   Claurst opens a full terminal UI powered by ratatui. You will see a message history area, an input box at the bottom, and a status bar showing the current model and session cost. Streaming responses appear in real time as Claude types.

   Keyboard shortcuts in the TUI:

   Key	Action
   Enter	Submit prompt
   Ctrl+C	Cancel streaming / clear input / quit
   Ctrl+D	Quit (when input is empty)
   Up / Down	Navigate input history
   Page Up / Page Down	Scroll message history
   F1 or ?	Toggle help overlay

4. Try your first prompt

   Type a prompt and press Enter. Claude will respond in the message area, and you will see tool calls appear as status messages while they run.

   > Explain what this project does based on the README
   

   Claude reads your project files, runs tools as needed, and streams its response directly in the terminal. Tool executions show a status message like Running Read... while in progress.

   For a non-interactive one-shot command (useful for scripts), pass the prompt directly:

   claude "List all TODO comments in the codebase"
   
5. Use a slash command

   Slash commands give you direct control over the session. Type / to start a command.

   > /help
   

   /help lists every available slash command with a short description. Use it any time you want to discover what’s available — there are 34 built-in commands covering session management, model switching, MCP server control, git workflows, and more.

   A few commands worth trying right away:

   Command	What it does
   /model	Show or change the active Claude model
   /cost	Display cumulative token cost for the session
   /compact	Manually compact the conversation to save context
   /plan	Enter plan mode (read-only — Claude proposes, you approve)
   /status	Show session metadata and current configuration
   /clear	Clear conversation history and start fresh

   See the Commands reference for the complete list.

Next steps

Installation

PATH setup, environment variables, feature flags, and system requirements.

CLAUDE.md configuration

Teach Claude about your project with a persistent project-level instruction file.

Tools reference

Every tool Claude can invoke: file ops, shell, web, agents, cron, and more.

MCP integration

Extend Claurst with external Model Context Protocol servers.