Dev Tools / Aider

Aider

by aider-chat

cli active free

An open-source AI pair programming CLI that works in any terminal, makes git-committed edits across multiple files, and supports over 100 LLM providers.

Aider is an open-source AI pair programming tool that runs in the terminal. The developer adds files to an editing session with natural-language instructions, and Aider applies the changes across multiple files and commits them to git automatically. It supports over 100 LLM providers through LiteLLM and is notable for consistently high scores on SWE-bench, the standard software engineering benchmark.

Key capabilities

Multi-file editing with git commits — Aider applies changes across all files the developer has added to the session and commits each logical change to git with an auto-generated commit message. The result is a clean, reviewable history rather than a blob of uncommitted edits.

Broad model support — Aider works with Claude (Anthropic), GPT-4o (OpenAI), Gemini, Llama, Mistral, DeepSeek, and many more through LiteLLM. Developers can use API keys for any provider or route requests through a local Ollama instance.

Repository map — Aider builds a tree-sitter-powered map of the entire repository (function and class signatures, call graph) that it passes to the model as context. This lets the model understand the full codebase structure even when only a subset of files is in the active session.

Architect mode — A two-model mode where one model proposes the change and a second (often a cheaper model) applies the edits. This reduces cost while maintaining quality for large refactors.

Lint and test integration — Aider can run the project’s linter and test suite after each edit and automatically iterate on failures until the tests pass.

Autonomy level

Level 3 (supervised agent): the developer reviews proposed changes and confirms before they are committed. Watch mode and --auto-commits can reduce the confirmation gates for trusted workflows.

Strengths

  • Top SWE-bench scores among open-source coding tools
  • Model-agnostic; works with virtually any LLM provider
  • Git-native workflow produces a clean, attributable history
  • Repository map makes it practical on large codebases

Limitations

  • Terminal only; no IDE visual experience
  • Requires the developer to manually add files to the session context
  • Architect mode doubles the API cost for each edit

Sources

Last verified June 12, 2026