AI pair programming that lives in git
Aider is a coding assistant that runs in your terminal and treats git as the system of record. You point it at a repo, describe a change, and it edits the files and commits the result with a sensible message. That last part is the whole design philosophy: every change the model makes becomes a real commit, so you review, diff, and undo with the git tools you already use rather than a bespoke history panel. If an edit goes wrong, git diff and git reset are your safety net.
It is one of the longer-running open-source coding tools, created in 2023 and sitting at 46,274 stars as of 2026-06 under the Aider-AI organization (it moved there from its original paul-gauthier/aider home). Its README reports 6.8M installs and 15B tokens processed per week, which, badge marketing aside, signals a large real-world user base.
What it does
The loop is conversational. You add files to the chat, ask for a change, and Aider sends the relevant context to your chosen model, applies the returned edits, and commits. It supports a wide range of models through their APIs (Claude, GPT, Gemini, DeepSeek, and local or OpenAI-compatible endpoints), and you switch models with a flag rather than a restart. It can also run linters and tests and feed failures back to the model to iterate, which is where the terminal-native design pays off in scripting.
Install
The README’s recommended path uses a dedicated installer so Aider runs in its own isolated environment:
python -m pip install aider-install
aider-installThen start it inside a git repo and name your model:
aider --model sonnet --api-key anthropic=<key>pipx install aider-chat and uv tool install aider-chat are common alternatives that achieve the same isolation. Check the official install docs if your platform needs anything special.
The repo map
Aider’s answer to “the model can’t see my whole codebase” is the repo map. It parses your project with tree-sitter across many languages and builds a compact map of the most relevant definitions and their relationships, then sends that instead of dumping entire files. On a medium project this is what lets it make coordinated edits without blowing the context window. It is also the feature most likely to strain on a very large repository, which the gotchas section gets into.
Where it fits, and where it doesn’t
Reach for Aider if you live in the terminal, want every AI change captured as a git commit, and want to move between models freely. It suits scripted and CI-adjacent workflows, and it shines on changes where a clean commit history matters for review. Its self-reported polyglot leaderboard, which it uses to rank how well models actually edit code, is a useful side artifact when you are choosing a model.
It is a weaker fit if you want an in-editor experience with inline diffs and a GUI, where an IDE extension is more natural. It also assumes a git repo; working outside one is possible but cuts against the grain of the tool.
aider versus the IDE agents
| Aider | cline | Continue | OpenHands | |
|---|---|---|---|---|
| Stars | 46,274 | 63,353 | 33,704 | 77,254 |
| Surface | terminal | VS Code, JetBrains, CLI | VS Code, JetBrains | Docker app, CLI |
| Shape | git-native pair programmer | IDE agent with checkpoints | embedded assistant | sandboxed agent |
| Review model | git commits and diffs | in-editor diffs and rollback | inline edits | sandbox plus diffs |
| Best at | clean commit-per-change history | reviewed edits across files | inline editor help | hands-off long tasks |
Counts are from GitHub as of June 2026. cline gives you the same bring-your-own-model freedom inside an IDE with rollback checkpoints. Continue was the embedded-assistant option and is now read-only. OpenHands goes fully autonomous in a sandbox. Aider’s distinct bet is that git itself is the best interface for reviewing AI changes.
The star curve
The star history climbs steadily over a long period rather than spiking, which matches a tool that built its reputation slowly among terminal-first developers and kept compounding as model quality improved. The recent slope is gentler, consistent with a mature project rather than a launch-week sensation.
What the issue tracker warns you about
Three patterns recur. First, the repo map can fail on very large codebases: the tracker carries multiple reports of errors thrown from the repo-map code on big repositories, so on a huge monorepo you may need to scope which directories Aider sees. Second, edit-format failures cost tokens: when the model’s diff does not apply cleanly Aider can fall back to rewriting whole files, which inflates spend, so watch your usage on a new model. Third, be careful what you add to context: files pulled in for reference are sent to the model and persist in the chat history, so do not add anything with secrets in it. A smaller quirk worth knowing is that commit-message generation can use a different model than your main one, which has surprised users expecting a single model setting.
Related
For the same model freedom inside an IDE, see cline. For the now read-only embedded assistant, see Continue. For a sandboxed autonomous agent, see OpenHands. Another terminal-friendly agent to compare is opencode. For what is trending, see LLM tooling, the daily digest, and the weekly report.
FAQ
Is Aider free? Yes, it is open source under Apache-2.0. You pay your model provider for token usage; there is no fee for the tool itself.
Aider vs cline, which should I use? Aider runs in the terminal and commits every change to git, which fits a CLI-centric, diff-driven review style. cline runs in your IDE with inline diffs and rollback checkpoints. Pick by where you want to review changes: git history or the editor.
Which model is most cost-effective with Aider? It depends on your work, and Aider’s own polyglot leaderboard is the honest place to compare editing quality across models. Frontier Claude and GPT models tend to apply edits most reliably; cheaper models can work but raise the rate of failed diffs and retries, which adds token cost.
How do I control Aider’s token cost? Add only the files you need to the chat, scope the repo map on large projects, and prefer a model with a high edit success rate so it retries less. Watch the usage Aider prints after each exchange.
Is Aider still maintained? Yes. It is developed under the Aider-AI organization, with the most recent commits as of 2026-05. Release cadence has slowed from its peak but the project remains active.