What is Aider?
Aider's real differentiator is not that it runs in the terminal. It is that every change is a git commit, automatically, and that single design choice is also its main tradeoff. Launch it inside a git repo, add files to the chat context, describe the change in plain English. Aider edits the files on disk and commits with a generated message. Your git history becomes the audit log of every AI-assisted change, which is exactly what you want in a reviewed codebase and exactly what gets noisy if you iterate in small steps.
Why the git-native model matters
A commit per change means review and rollback are ordinary git operations: git diff HEAD~1, git reset, git revert. No proprietary undo stack, no editor-specific history to learn. The cost is commit volume. An exploratory session can produce a dozen tiny commits, so teams using Aider seriously squash before merge or run it on a scratch branch. Know that going in and it is a non-issue; ignore it and your history gets noisy fast.
The other load-bearing piece is the repo map. Aider builds a structural map of the codebase so the model can reason about symbols in files you have not added to the chat. This is what lets it make a coordinated multi-file change instead of editing one file blind to its callers. The map is also why Aider degrades on very large repos where the map itself competes for context budget; scope the chat to the relevant files and it holds up.
Key capabilities
- Git-native workflow: every edit is a commit, so review and rollback are normal git, not a bespoke history UI
- Repo-map awareness: a structural map lets the model reason about files outside the immediate chat context
- Multi-file edits: coordinated changes across several files in one turn, anchored by the repo map
- Voice coding: optional speech-to-prompt for hands-free pairing
- Model-agnostic: Anthropic, OpenAI, DeepSeek, and local models via Ollama; you pay the provider directly
Who should use it, and who should not
Use it if you live in the terminal, drive by keyboard, and want AI changes that land as reviewable commits with no GUI in the loop. It is also strong as a scripted, CI-adjacent tool for repeatable changes across many repos, precisely because the output is plain git.
Skip it if your team is not comfortable in git internals, or if you want an agent that explores ambiguous tasks with a rich interactive UI. A GUI editor like Cursor is the better call there. The decision is not terminal-versus-GUI taste. It is whether "every change is a commit" fits how your team reviews code.
Pricing
Free and open-source. There is no Aider subscription; your only cost is the per-token charge from whichever model provider you point it at, so spend tracks your model choice, not Aider.