Authority AI Editorial
Aider: pair programming in the terminal
Aider is an open-source command-line tool that lets you edit a codebase by talking to an LLM in your terminal. You describe a change in plain language, Aider figures out which files to touch, writes the edit, and commits it to git with a message it generates. It is built for developers who want an agent that lives in the terminal and treats version control as part of the workflow, rather than a chat window bolted onto an editor.
What it does
Aider reads your project, builds a map of the codebase so it can find relevant files in large repositories, and applies edits directly to disk. That repository map is built with a graph ranking algorithm: Aider treats each source file as a node in a graph, connects files that depend on each other, and surfaces the classes and functions referenced most often elsewhere in the code. The map is sized to a token budget, 1,000 tokens by default, and Aider's documentation says it expands or shrinks that budget as a chat session goes on, so a large repository does not overwhelm the model's context window. Each change Aider makes is committed to git automatically, so every step of a session has a rollback point you can inspect with ordinary git tools. It supports more than 100 programming languages and can take in images, web pages, and voice input as part of a request. It can also run your linter and test suite after a change and feed failures back into the next round of edits, so a broken test becomes part of the next prompt rather than something you have to notice yourself.
How you use it
You install Aider with pip install aider-install followed by aider-install, then run it from inside a project directory with a model and API key, for example aider --model sonnet --api-key anthropic=<key>. From there you chat with it about the change you want, describing it the way you would to a colleague. Aider decides which files need editing based on your request and its repository map, shows you the diff before or as it applies the change, and commits once you accept it, writing its own commit message. It also has a watch mode that lets you leave comments in your editor for Aider to act on, so you can stay inside VS Code or another IDE and use Aider as the execution layer underneath it rather than switching to a separate terminal window for every request. For projects that already have a test suite, you can tell Aider which command to run after each edit, and it will read the failures and try again without you copying error output back in by hand.
What to know before adopting it
Aider is licensed under Apache 2.0 and is written in Python. It has no server component: it runs locally against whichever model you point it at, so the cost is whatever your model provider charges per token, not a subscription to Aider itself. It works with Claude, GPT, Gemini, DeepSeek, and other hosted models, and it also supports local models for people who want to keep code off third-party infrastructure. According to the project's own release notes, Aider wrote 88 percent of the code checked into its v0.86.0 release, which the maintainers use as a running measure of how far the tool can extend itself. That is a claim from Aider's own changelog, not an independent audit, so read it as a self-reported metric rather than a benchmark.
Where it fits next to similar tools
Aider sits at the terminal end of the coding-agent spectrum: no GUI, no browser panel, just a REPL that edits files and commits them. That makes it a good fit for people who already live in a terminal and want an agent that treats git history as the audit trail, and a plainer fit than an IDE-embedded agent for scripting or CI use. If you would rather hand your coding agent a written method than retype the same instructions, the AI Ultimate Skill Collection is 3,029 ready-made skills in 12 packs, each a written procedure your agent follows step by step.