Aider is an open-source AI coding assistant that runs in the terminal, giving you access to powerful AI-assisted coding without switching to a new IDE or paying for a proprietary tool. It connects to your existing code editor and git repository, reads the files you specify, and makes changes based on natural language instructions. Aider has built a strong following among developers who prefer terminal-centric workflows, want full control over which AI model they use, and value open-source transparency over polished commercial products. This guide covers installation, the core workflow, and the practices that make Aider effective.
How Aider Works
Aider runs as a command-line program in your terminal. You start it in a git repository, tell it which files are relevant to the task, give it a natural language instruction, and it makes the necessary code changes. The changes are written directly to your files and optionally committed to git automatically. Aider maintains a conversation that accumulates context across multiple instructions in a session — you can build on previous changes without re-explaining the context. The tool reads the specified files in full and sends them to the AI model along with your instruction; the model returns the modified files; Aider applies the changes. It is a simple, transparent loop with no hidden orchestration and no proprietary cloud infrastructure involved beyond the AI model API itself.
Installation and Setup
Install Aider with pip:
pip install aider-chatAider requires an API key for whichever model you want to use. Set it as an environment variable:
export ANTHROPIC_API_KEY=your-key-here # for Claude
export OPENAI_API_KEY=your-key-here # for GPT-4oStart Aider in a git repository:
cd your-project
aiderAider opens a chat interface in the terminal. Add the files relevant to your task with the /add command, then describe what you want in plain English. The first session typically takes a few minutes to learn the interface; by the second or third session the workflow feels natural.
The Core Workflow
A typical Aider session follows a consistent pattern. Open Aider in the project directory. Add the files relevant to the current task using /add filename.py or /add src/ for a directory. Type your instruction: “Add input validation to the create_user function — email must be a valid email address and username must be 3–20 alphanumeric characters.” Aider sends the files and instruction to the model, receives the modified code, and applies the changes. Review the diff displayed in the terminal. If the changes look correct, they are already in your files (and committed to git if auto-commit is enabled). If not, type a follow-up instruction to correct the issue. This loop — instruction, review, iterate — is the core of effective Aider use.
Figure 1 — Aider workflow and key commands
Choosing the Right Model
Aider supports a wide range of AI models, and the model choice significantly affects output quality. Claude Sonnet is the most popular choice among Aider users for general coding tasks — it produces accurate, well-structured code changes and handles multi-file edits reliably. GPT-4o is a strong alternative with slightly different strengths (sometimes better at certain Python patterns). For users who want to avoid API costs entirely, Aider supports local models via Ollama — the quality is lower than frontier models but sufficient for many tasks and completely private. Aider’s leaderboard (available at aider.chat/docs/leaderboards) shows benchmark performance of various models on real coding tasks specifically tested with Aider, which is more relevant than general coding benchmarks for choosing the right model for this workflow.
Git Integration and Auto-Commits
Aider’s git integration is one of its strongest features. By default, it auto-commits each change it makes with a descriptive commit message generated from your instruction. This gives you a clean, granular commit history where every AI-made change is separately reversible. If a change is wrong, /undo reverses the last commit instantly. You can review the full git log to see exactly what Aider did across a session. If you prefer to control commits manually, set auto-commits = false in the Aider config — Aider will still make the changes to your files but leave committing to you. For most workflows, auto-commits are recommended: they make the session auditable, keep related changes grouped correctly, and eliminate the overhead of manually committing every incremental change.
Context Management: Adding the Right Files
Aider’s quality depends heavily on having the right files in context. Too few files and the model lacks the information to generate correct changes. Too many and the context window fills, causing the model to lose track of earlier content and increasing API cost. The practical rule: add the files that the AI needs to read to understand the current code structure, and the files that need to be changed. If you are adding a new feature to a service layer, add the service file, the relevant repository file, and any models the service uses. Do not add the entire project. Use /drop to remove files that were relevant earlier in the session but are not needed for the current task. Aider also supports a read-only mode for files: /read file.py adds a file for context without allowing the AI to edit it, useful for providing reference implementations or interfaces without risking unintended modifications.
Using Aider for Test-Driven Development
Aider works particularly well for test-driven development workflows. The pattern: write a failing test manually (or ask Aider to write the test first), run the test to confirm it fails, then ask Aider to implement the code that makes the test pass. Aider can run your test suite with /run pytest or /run npm test, observe the output in the conversation, and iterate on the implementation until the tests pass — all within a single session. This combination of TDD structure with AI implementation is one of the most reliable patterns for producing correct, well-tested code with minimal human implementation effort. The test acts as the specification, the AI handles the implementation, and the test result provides objective verification.
Aider Architect Mode
Aider has an architect mode (enabled with --architect) that uses two models: a more capable “architect” model for planning and reasoning about the approach, and a cheaper “editor” model for actually making the code changes. This is particularly useful for complex tasks where the planning requires frontier model capability but the implementation is straightforward enough for a smaller model to execute. The architect model describes the changes needed in natural language; the editor model applies them to the code. In practice, this produces better results on complex multi-file refactors while reducing the API cost of the editing work. The default configuration uses Claude Sonnet as architect and a smaller Claude model as editor, but both are configurable.
When Aider Is the Right Tool
Aider is the best fit for developers who prefer terminal-centric workflows and want the flexibility to use any AI model without being locked into a specific IDE. It is particularly strong for Python projects (where its test integration is most mature), for developers already comfortable with git who want tight version control of AI changes, and for workflows where privacy matters and a local model via Ollama is preferred over cloud APIs. It is less suited to frontend-heavy work where visual feedback from a browser preview is important, or for developers who want a polished GUI rather than a terminal interface. The open-source nature means the community moves quickly — checking the Aider GitHub repository for recent releases often reveals useful new features that have not yet made it into guides.
Aider occupies a distinct position in the AI coding tool landscape: open-source, terminal-native, model-agnostic, and tightly integrated with git. For developers who want maximum control and transparency, it is one of the most flexible AI coding assistants available. The learning curve is modest — a few sessions to learn the commands, a few more to develop intuitions about which files to add and how to phrase instructions effectively. After that, it becomes a fast, low-friction tool for AI-assisted coding that fits naturally into existing terminal-based development workflows.
Aider Configuration File
Aider can be configured persistently via a .aider.conf.yml file in the project root or your home directory. Useful settings to configure: model sets the default AI model so you do not specify it on every startup. auto-commits controls whether Aider commits automatically. read specifies files to always include in read-only context — useful for architecture documentation, interface definitions, or any file the AI should know about but never modify. lint-cmd runs a linter after each change so the AI can see and fix lint errors automatically. test-cmd specifies the test command Aider runs after each change in test mode. Setting these once in the config file eliminates repetitive session setup and makes Aider’s behaviour consistent across sessions without any manual configuration overhead.