Code CLI Overview

Get started in 30 seconds

Prerequisites:

• An Autohand account or API key from your preferred provider:
  • OpenRouter - Access 200+ models including Claude, GPT-4, Llama (recommended)
  • Ollama - Run models locally for privacy and offline use
  • llama.cpp - Maximum performance with GGUF models
  • MLX - Native Apple Silicon acceleration (M1/M2/M3/M4)

Install Autohand CLI:

curl -fsSL https://autohand.ai/install.sh | sh

brew tap autohandai/code && brew install autohand-code

npm i -g autohand-cli

iwr -useb https://autohand.ai/install.ps1 | iex

Start a session:

cd your-project
autohand

You will be prompted to log in on first use. That's it. Continue with Configuration

Quick start

Here are the most common ways to use Autohand Code from the terminal:

# Start an interactive session
autohand

# Use pipe mode for Unix composability
git diff | autohand 'explain these changes'

# Start with a specific model
autohand --model claude-4-sonnet

# Run in plan mode (read-only exploration first)
autohand --plan

# Auto-mode with safety limits
autohand --automode --max-iterations 10 --max-cost 5

# Enable extended thinking for deep reasoning
autohand --thinking

Modes of operation

Autohand Code supports several modes to fit different workflows:

• Interactive mode (default) - Start a conversation, give instructions, review changes as they happen.
• Pipe mode - Feed data via stdin for Unix composability. echo "explain this" | autohand outputs structured results to stdout.
• Plan mode - Toggle with Shift+Tab. The agent explores your codebase with read-only tools first, then executes only after your approval.
• Auto-mode - Autonomous development loops where the agent plans, codes, tests, and iterates. Built-in guardrails limit iterations and cost.
• YOLO mode - Use --yolo for granular auto-approve patterns. Skip confirmation prompts for commands you trust.
• Headless mode - For CI/CD pipelines and automation. No interactive prompts, pure programmatic control.

Key features

• Agent Teams - Orchestrate multiple agents working in parallel on different parts of a task. Split large projects across specialized workers.
• Skills System - Modular workflow packages you can install, share, and create. Auto-skill generation turns repeated patterns into reusable skills.
• MCP Support - Connect external tools, databases, and APIs through the Model Context Protocol.
• Extended Thinking - Use --thinking for deep reasoning on hard problems. The agent shows its chain of thought before acting.
• Session History - Browse and resume past sessions with /history. Pick up where you left off.
• Context Compaction - Automatic conversation summarization keeps the agent effective during long sessions.
• Custom System Prompts - Shape agent behavior with --sys-prompt and --append-sys-prompt for project-specific instructions.
• 50+ slash commands - Quick actions for common tasks like /commit, /review, /test, and more.
• 50+ built-in tools - File editing, search, git operations, web fetch, and more available out of the box.
• 15+ language localizations - Use Autohand in your preferred language.
• IDE detection - The /ide command detects your editor and opens files directly in it.

See it in action

Watch Autohand Code handle real tasks from your terminal. Each demo runs a single command and shows the full agent workflow.

What Autohand Code does for you

• Build features from descriptions: Tell Autohand what you want to build in plain English. It plans, writes the code, and verifies it works.
• Debug and fix issues: Describe a bug or paste an error message. Autohand analyzes your codebase, finds the problem, and implements a fix.
• Navigate any codebase: Ask anything about your project and get a clear answer. Autohand maintains awareness of your entire project structure and can pull up-to-date information from the web.
• Automate tedious tasks: Fix lint issues, resolve merge conflicts, write release notes, and migrate legacy code in a single command from your terminal or automatically in CI.

Why developers love Autohand Code

• Works in your terminal: Not another chat window. Not another IDE. Autohand meets you where you already work, with the tools you already use.
• Takes action: Autohand edits files, runs commands, creates commits, and delegates work to agent teams for parallel execution.
• Unix philosophy: Composable and scriptable. tail -f app.log | autohand -p "alert me if you see anomalies" works. Your CI can run autohand -p "translate new strings and raise a PR".
• Provider flexibility: Connect to OpenRouter, Ollama, OpenAI, or run local models with llama.cpp and MLX. Use the provider that fits your needs and budget.

Detailed documentation

Additional resources

Frequently asked questions