Aider is an open-source AI pair programming tool that runs entirely in the terminal, enabling developers to edit code in real Git repositories through natural language conversation with large language models. With 41,500 GitHub stars and over 5.3 million PyPI installations, it has become one of the most widely adopted AI coding assistants in the open-source ecosystem, processing over 15 billion tokens weekly across its user base. ## How Aider Works Unlike browser-based AI coding tools that operate on isolated code snippets, Aider works directly within your existing Git repository. When you start Aider in a project directory, it builds a comprehensive map of your codebase's structure, identifying files, functions, classes, and their relationships. This repository map serves as context that helps the LLM understand the broader architecture when making changes to specific files. You describe what you want in natural language: "Add rate limiting to the API endpoints" or "Refactor the authentication module to use JWT tokens." Aider translates these requests into precise code edits, applies them to the actual files, and automatically creates Git commits with descriptive messages. If the changes break tests or trigger linter warnings, Aider can detect and fix those issues automatically. ## Multi-Model Support Aider connects to virtually any LLM through a flexible model configuration system. It works optimally with: | Model | Strength | |-------|----------| | Claude 3.7 Sonnet | Best overall coding performance | | DeepSeek R1 / Chat V3 | Cost-effective reasoning and generation | | OpenAI o1 / o3-mini | Strong reasoning with chain-of-thought | | GPT-4o | Fast general-purpose coding | | Local models (Ollama) | Privacy-preserving offline use | The `/model` command switches between models mid-session, and Aider automatically adjusts its edit format to match each model's strengths. The `/think-tokens` command controls the thinking token budget for reasoning models, accepting human-readable values like `8k` or `0.5M`. ## Repository Map Technology Aider's repository map is its core differentiator. Using tree-sitter parsing across 100+ programming languages, it builds an abstract syntax tree of the entire codebase, extracting function signatures, class hierarchies, import relationships, and module boundaries. This map is compressed and included in the LLM context, allowing the model to understand how a change in one file might affect others, even when those files are not explicitly included in the conversation. The map updates dynamically as files change, and Aider intelligently selects which portions to include based on relevance to the current task. This enables effective editing in large repositories where the full codebase far exceeds any model's context window. ## Git Integration Every edit Aider makes is automatically committed to Git with a meaningful commit message. This provides a complete audit trail of AI-assisted changes and makes it easy to review, revert, or cherry-pick individual modifications. Aider respects existing Git workflows: it works with branches, handles merge conflicts, and never force-pushes or modifies history. ## IDE and Editor Integration While Aider is terminal-native, it integrates with popular editors through multiple mechanisms. Developers can add special comments in their code (e.g., `# ai: refactor this function`) that Aider picks up and acts on. The Ctrl-X Ctrl-E keybinding opens the current input in an external editor for complex prompts. Aider also supports a web chat interface that streamlines copy-paste workflows with any LLM's browser UI. ## Voice and Multimodal Input Aider accepts voice input for hands-free coding requests, translating spoken descriptions into code changes. It also processes images (screenshots, mockups, diagrams) and web page URLs as context, allowing developers to say "make this page look like this mockup" with an attached image. ## Automated Quality Assurance After every code change, Aider can automatically run the project's linter and test suite. If issues are detected, it attempts to fix them in a follow-up edit cycle. This creates a tight feedback loop where code quality is maintained without manual intervention. ## Performance and Scale Aider has been battle-tested on repositories ranging from small scripts to large monorepos with hundreds of thousands of lines of code. Its SWE-bench Verified score of 26.3% (as of mid-2025) demonstrated strong real-world bug-fixing capability on actual open-source projects. ## Community and Ecosystem With 41,500 stars, 4,000 forks, 13,102 commits, and active development (last updated March 3, 2026), Aider maintains one of the most active open-source AI coding communities. The project is Apache 2.0 licensed and developed by Paul Gauthier and contributors. It has received praise from notable developers including Eric S. Raymond, who described it as life-changing.