Aider

Aider is a command-line tool that lets a large language model edit code directly in your local Git repository. It is open source, written in Python, and installed with a package manager. You run it inside a project, tell it what you want, and it reads the relevant files, produces edits, applies them to disk and commits the result with a descriptive message. There is no proprietary editor to adopt, no account to create and no subscription to pay to the tool itself. That last point is the structural difference from most commercial AI coding products, and it shapes everything else about how Aider is used.

The central technical idea in Aider is the repository map. Language models have finite context windows, and a real codebase does not fit in one. Naive tools either stuff the whole project into the prompt and pay for the tokens, or rely on the user to attach files manually. Aider instead builds a compressed map of the repository using tree-sitter parsing: it extracts the signatures of classes, functions and methods across the codebase, ranks them by relevance to the files currently being edited using a graph-based algorithm, and includes as much of that skeleton as fits the token budget. The model therefore understands the shape of the project, the names and signatures it might need to call, and where things live, without reading every line. As you add files to the chat, the map adjusts. This is what makes Aider usable on medium and large codebases rather than only on toy projects.

Editing works in several modes. The default asks the model to produce a concise search-and-replace style diff, which is token-efficient and works well with strong models. Other modes ask for a whole updated file, or for a unified diff format, and the right choice depends on the model you are using and the size of the change. Aider validates that edits applied cleanly and reports when they did not. Every change is committed to Git automatically with a message that summarises what was done, which means you can review, revert or cherry-pick AI work with ordinary Git commands. A sensible workflow is a dedicated branch for AI-assisted changes, reviewed and squashed before merging.

Model support is deliberately broad. Aider works with hosted APIs from OpenAI, Anthropic, Google, DeepSeek, Mistral, OpenRouter, Groq and others, and it also works with locally served open-weight models through Ollama, llama.cpp, or any OpenAI-compatible endpoint. You supply your own API key, so cost is whatever your provider charges for the tokens consumed, and there is no middleman markup. This matters for two reasons. First, it lets you choose a model based on current quality and price rather than on what a vendor has bundled. Second, it allows fully offline operation for codebases that cannot leave the machine, which is a hard requirement in some regulated and defence-adjacent environments.

The interface is a terminal chat. You can add and drop files from the conversation, run shell commands with the output fed back to the model, ask for an explanation before making a change, and use voice input to dictate requests. There is a browser-based UI available as an option, and Aider can integrate with editor workflows by watching files, but the terminal remains the primary surface. Architect mode splits work between a stronger model that plans the change and a cheaper editor model that applies it, which can improve quality per dollar. Watch mode and scripting support let you drive Aider from CI or from a file of instructions, which opens batch refactoring use cases.

Against competitors, Aider occupies the open, terminal-native corner of the market. GitHub Copilot lives inside editors and is excellent at inline completion and increasingly at multi-file agent tasks, but it is a subscription tied to a specific vendor's model routing. Cursor is a full editor fork of VS Code with a very polished chat and agent experience, and it is the choice for people who want a graphical tool; it is closed source and subscription-based. Claude Code, Gemini CLI and the OpenAI Codex CLI are the closest direct rivals: also terminal agents, also able to run commands and edit files, but tied to a single provider's models and, in most cases, to that provider's account and billing. Continue is an open-source editor extension that shares Aider's model-agnostic philosophy but lives in the IDE rather than the shell. Cline and Roo Code are open-source editor agents with similar autonomy. Aider's distinguishing combination is model agnosticism, a mature repo map, tight Git discipline and a genuinely free licence.

Pricing is straightforward. Aider is free software under a permissive open-source licence, with the source published and contributions accepted. You pay only for model inference, which can be nothing at all if you run a local model. Real-world token spend depends heavily on how many files you put in the chat and how large the repository map is; a heavy day of agentic editing against a premium hosted model can cost a few dollars, while local models cost only electricity and patience. There are no seats, no tiers and no enterprise contract, which removes procurement friction entirely but also means there is no vendor support organisation to escalate to.

The limitations are real. Terminal-only interaction has a learning curve and no visual diff review beyond what Git and your editor provide, so people who want to click through suggested changes will find Cursor or an IDE extension more comfortable. Quality is entirely dependent on the model you connect; Aider does not improve a weak model, and cheap models produce edits that fail to apply or that misunderstand intent. Large multi-file refactors can still go wrong, and the automatic commit habit means you need branch discipline to avoid polluting history. Token costs on hosted models scale with repository size, and the repo map, while efficient, is not free. There is no centralised team administration, no audit log service and no compliance paperwork, so an enterprise that needs those has to build them around the tool. Finally, as an open-source project, its maintenance depends on its contributors, and features move quickly, which occasionally means breaking changes between releases.

Aider fits experienced developers who live in a terminal, teams that want to control which model touches their code, organisations with restrictions on sending source to a specific vendor, and anyone doing scripted or batch refactoring across a repository. It fits solo engineers and small teams who want AI assistance without adding another subscription. It does not fit developers who want a graphical editor experience, non-technical users, or enterprises that require vendor support contracts, centralised policy enforcement and audit trails out of the box.

In the closing comparison, choose Aider when you value openness, model choice and Git-native discipline over interface polish. Choose Cursor when you want the best integrated graphical experience and are comfortable with a subscription. Choose GitHub Copilot when you want the lowest-friction path inside the editor you already use with institutional support behind it. Choose Continue when you want open source and model agnosticism but prefer to stay inside VS Code or JetBrains. Choose Claude Code or a similar provider-specific CLI when you are already committed to that provider and want the tightest integration with its strongest models. For a command-line-first engineer with strong opinions about which model to use, Aider is currently one of the best tools available.