The Terminal Pair Programmer
Aider is the open-source AI coding assistant that predates the current agentic wave and still holds its ground. It lives in your terminal, talks to any LLM API you give it, and edits files directly in your repo — committing each change to git as it goes. No subscription, no telemetry, no vendor lock-in. Just a Python package and your API key.
In a market dominated by funded products like Cursor and Claude Code, Aider’s appeal is radical simplicity and total ownership. You control the model, the context, the edit strategy, and the commit history.
The Git-Native Workflow
Aider’s defining feature is that every edit is a git commit. Ask it to “add input validation to the login function” and it edits the file, stages it, and commits with a descriptive message. If the edit is wrong, git revert undoes it. This is the safest edit model of any AI coding tool because your version control system is the undo stack — no proprietary history, no hidden state.
The workflow is conversational. You describe what you want, Aider shows which files it plans to edit, applies the changes, and commits. You can ask for refinements, and each refinement is another commit. The result is a clean, reviewable git log of AI-assisted changes.
Multi-Model Support and Routing
Aider supports any model with an API — OpenAI, Anthropic, Google, local models via Ollama. The standout feature is model routing: use a cheap, fast model (like GPT-4o-mini) for simple edits and a strong model (like Claude Opus) for complex refactors. This lets you optimize cost without sacrificing quality on hard tasks.
| Feature | Aider | Claude Code | Codex CLI |
|---|---|---|---|
| Open source | Yes (Apache 2.0) | No | Yes (Apache 2.0) |
| Git-native commits | Every edit | Manual | Manual |
| Multi-model | Any API model | Claude only | OpenAI models |
| Agentic loop | No | Yes | Yes |
| IDE integration | None | Terminal + IDE | Terminal |
Edit Formats
Aider’s most confusing aspect for newcomers is the edit format system. Different models work better with different edit strategies — whole-file replacement for small files, search-and-replace blocks for large files, unified diffs for precise edits. Aider auto-selects the best format per model, but advanced users can override it. Getting this right matters: the wrong format on a large file produces broken diffs that fail to apply.
Where It Lacks
Aider is a pair programmer, not an agent. It edits files but does not run your test suite, read failures, and iterate. When a test breaks, you paste the output back into the conversation and ask for a fix — same as Copilot Chat, just in the terminal. The agentic loop that makes Claude Code and Codex CLI powerful is absent. For teams that want “set a task and walk away,” Aider is the wrong tool. For engineers who want tight, reviewable, git-native edits, it is excellent.
Verdict
Aider is the tool for engineers who want AI assistance without giving up control or paying a subscription. It is not the most powerful — it lacks the agentic loop — but it is the most transparent. Every edit is a commit, every commit is yours, and no vendor can deprecate your workflow. If you live in the terminal and value open source, Aider belongs alongside Codex CLI in your toolchain. Pair it with a strong model for complex work and a cheap model for everything else, and you have a cost-effective, durable AI coding setup.