Aider AI: Complete Guide to the Git-Aware CLI Coding Agent

What Is Aider?

Aider is a CLI coding agent built by Paul Gauthier. Its defining feature is git-awareness: every change Aider makes is automatically committed with a descriptive commit message. This gives you a clean, auditable history of all AI edits — easily reversible with git revert or Aider's own /undo command.

Aider works with any language and supports multi-file edits. You specify which files are relevant, describe the change in plain English, and Aider applies the diff and commits. Key capabilities:

• Multi-file edits with automatic git commits
• Support for 100+ AI models (Claude, GPT-4o, DeepSeek, Gemini, Ollama)
• Architect mode: separate planning model + editor model
• Auto-run tests after each change (--auto-test)
• Voice input support
• Linting and auto-fix on each edit

Aider is MIT-licensed, free to use, and you bring your own API key. Visit the aider.chat website for documentation.

Installation

Aider requires Python 3.9+ and an API key for your chosen model provider.

# Install via pip
pip install aider-chat

# Or via pipx (recommended for isolation)
pipx install aider-chat

# Set API key (Anthropic example)
export ANTHROPIC_API_KEY=sk-ant-...

# Navigate to a git repo and start
cd my-project
aider

# Start with a specific model
aider --model claude/claude-sonnet-4-6

# Start with OpenAI
aider --model gpt-4o

Key Slash Commands

Aider's interactive session uses slash commands to manage context and control the agent:

Architect Mode

Architect mode is Aider's most powerful feature for complex tasks. It separates thinking from writing:

• Architect model (e.g., claude-opus-4) — plans the approach, reasons about the codebase, decides what to change
• Editor model (e.g., claude-haiku-4-5) — takes the architect's plan and writes the actual code diffs

aider --architect \
  --model claude/claude-opus-4 \
  --editor-model claude/claude-haiku-4-5

Architect mode typically produces better results on complex refactors while keeping editing costs low. The architect's tokens cost more, but the editor's are cheap — net cost is often similar to running Sonnet alone.

Configuring .aider.conf.yml

Avoid typing flags every session by creating a .aider.conf.yml in your project root or home directory:

# .aider.conf.yml example
model: claude/claude-sonnet-4-6
auto-commits: true
auto-test: pytest
editor-model: claude/claude-haiku-4-5

Other useful config options include auto-commits: false (disable auto-commit, review first), read: CONVENTIONS.md (always read a context file), and dark-mode: true for terminal color scheme.

SWE-bench Performance

SWE-bench measures how well AI agents fix real GitHub issues. Aider consistently places among the top open-source tools:

Scores change frequently as models improve. See aider.chat/docs/leaderboards for the current benchmark results.

Aider vs Claude Code vs Cursor

Use Aider when you want terminal-native, git-committed AI edits with full model freedom. For more context, see our Claude Code guide and Cursor guide.

Power User Tips

• Use --auto-test: Run aider --auto-test pytest to automatically run your test suite after every change. If tests fail, Aider retries the fix automatically.
• Keep context tight: Only /add files that are directly relevant to the current task. Adding too many files inflates the context window and increases cost without improving results.
• Use /run for feedback loops: After Aider makes a change, use /run python myfile.py to show the output back to Aider for immediate iteration.
• Read-only context: Use aider --read README.md to provide context files that Aider can read but not edit.
• Monitor API uptime: Aider stops working when your AI provider is down. Track Anthropic status on Prismix to know immediately.