Aider Chat: Pair Program with LLMs

Tool snapshot
Best fitCoding · Productivity · Image Generation
In one lineAider provides terminal-based AI pair programming in Git repositories, with repository context, model connections, and testing support.

Aider is an AI pair-programming tool for working in a terminal and editing a local Git repository. You can use it for an existing codebase or begin a new project, choosing a compatible model connection for the coding work.

Repository context for edits

Aider sends a repository map with each change request. The repository-map documentation describes a compact view of files and important symbols, including classes and function signatures. It ranks relevant portions of that map to fit the available context budget.

That context helps the model identify related code, but it is a summary rather than a complete copy of every file. Aider can offer to add specific files to the chat when more detail is needed. For a proposed change, provide the relevant behavior and identify any constraints the edit must preserve.

Git, tests, and interaction

Aider can commit code changes to Git. Its homepage describes using familiar Git tools to inspect or undo those edits, and it advertises linting and testing after changes. The documentation also covers chat modes, coding conventions, and configuration-file editing.

You can supply images or web pages for context, and the publisher lists voice-to-code interaction. These options may help explain a user-interface change or describe a bug, although the resulting patch still needs code review and suitable tests.

Model setup and practical limits

Aider can use local models through Ollama. Its model guide also describes cloud providers and OpenAI-compatible APIs. The guide cautions that less capable models can struggle to return edits in a form Aider can apply.

Choose the connection based on your data handling requirements and the model's coding performance. Start with a small change on a branch, inspect the diff, and run the project's meaningful checks. Franklin has not benchmarked these model combinations; provider costs and available models depend on the selected service.