The Problem
Developers working with LLMs in the terminal face a fundamental gap: chat interfaces don't understand codebases, and code editors don't understand chat. Aider closes that gap by giving an LLM direct access to a repository—it can read files, propose edits, and commit changes without leaving the terminal. The pain point is real: copying code between a chat window and an editor is slow, error-prone, and loses the context the model needs to make correct changes.
What This Does
Aider is a Python CLI that pairs a developer with an LLM inside a Git repository. The core logic lives in aider/main.py (entry point) and aider/coders/basecoder.py, which orchestrates the edit loop. The aider/coders/ directory contains specialized implementations—editblockcoder.py for targeted edits, wholefilecoder.py for full-file rewrites, udiffcoder.py for unified diffs, and architectcoder.py for a two-model planning/execution split. Each coder has a corresponding *prompts.py file with the system prompts that define its behavior.
The repo map feature (aider/repomap.py) builds a tree-sitter–based index of the codebase so the model can locate relevant symbols across hundreds of files. Git integration is handled through aider/commands.py and aider/repo.py, which manage automatic commits and repository state. The aider/queries/tree-sitter-language-pack/ directory contains syntax queries for 30+ languages, enabling the codebase map to work broadly.
How To Use It
Setup: The repo is a Python package with no pyproject.toml or setup.py at the root—only a Gemfile for the Jekyll website under aider/website/. The README indicates installation via pip install aider-chat from PyPI. For development, you'd install from source with pip install -e ., though no lockfile exists to pin dependencies.
Configuration: API keys are set as environment variables (e.g., ANTHROPICAPIKEY, OPENAIAPIKEY). Model selection and provider routing happen via aider/models.py and aider/llm.py. The aider/resources/model-metadata.json and aider/resources/model-settings.yml files define model capabilities and defaults.
Running it: The CLI entry point is aider/main.py, invoked as aider after installation. The README shows the basic workflow: aider launches an interactive session where you describe changes in natural language, and the tool edits files and creates commits.
pip install aider-chat export ANTHROPICAPIKEY=your-key-here cd your-repo aider "Add a rate limiter to the API client"
Real-World Use
A typical session: you're working on a Django app with a bug in views.py. You run aider, type "Fix the off-by-one error in the pagination logic in views.py." The tool maps the repo, locates the relevant code, proposes an edit, and commits it with a message like "Fix off-by-one error in pagination." You review the diff with git diff HEAD~1, and if it's wrong, git revert undoes it. The aider/watch.py module extends this to watch for code comments—you add # aider: refactor this function and it acts without you touching the terminal.
Code Health & Issues
Med - No Python dependency lockfile (requirements.txt or pyproject.lock) at the root—reproducible installs are not guaranteed. The only lockfile is aider/website/Gemfile.lock for the Jekyll site. Med - Only 4 test files for 80 Python files. The .github/workflows/ directory shows CI (Ubuntu, Windows, pre-commit), but test coverage appears thin for a project of this complexity. Low - The aider/website/ directory mixes a Jekyll site with Python application code, which complicates the repo layout and dependency management. Low - aider/gui.py and aider/voice.py suggest features that may be less battle-tested than the core CLI path.
The project has solid CI coverage across OSes, a license, and active documentation. The tree-sitter query files are extensive and well organized.
The Bottom Line
Aider is a mature, practical tool for developers who want LLM assistance without leaving the terminal. It's genuinely useful for solo developers and small teams working in Git repos, and the repo map feature makes it viable on larger codebases. The lack of a Python lockfile and thin test suite are real concerns for production adoption, but for a developer tool, the core value proposition is strong and well-executed.