What Is Aider? Open-Source AI Pair Programming

Aider is an open-source AI pair programmer that runs in your terminal and edits the code in your git repository. You pick a model and add files to the chat. Then you describe the change. Aider writes the edit and commits it with a descriptive message. Its polyglot benchmark is a common reference for comparing models on code edits. If you want an AI coding tool that stays in the terminal and leaves a clean git history, this guide shows how Aider works and how to install it. It also shows what Aider forgets between sessions and how active the project is in late 2026.

What is Aider, in one sentence?

Aider is an Apache-2.0 command-line tool that pairs you with an LLM to edit code in your local git repository and commits each change automatically.

The GitHub repository describes it as "AI pair programming in your terminal." It had about 49.8k stars on October 28, 2026. The PyPI page lists Paul Gauthier as the maintainer. By its own count on the Aider homepage, Aider has 6.8 million PyPI installs and processes 15 billion tokens a week.

How does Aider differ from IDE agents like Cline?

Aider works from the terminal and treats git as its undo system. Editor agents such as Cline work inside VS Code and ask you to approve each step.

Sources: Aider's git docs and lint and test docs. Model support is on the model docs page.

How does Aider work?

Aider starts with a repo map. The repo map docs explain that it lists the files in your repository with their key classes and functions. It uses a graph ranking algorithm, with files as nodes and dependencies as edges, to include only the identifiers other code references most. The map is built with tree-sitter, per a 2023 post on its design. Its default budget is 1,000 tokens, enough to sketch the codebase for the model without sending every file.

You then add the files you want changed with /add. Files you only want the model to read go in with /read-only.

Aider asks the model to reply in an edit format. The edit formats page describes several. The whole format returns the full file. The diff format returns search-and-replace blocks. Aider picks a default per model, because some models are more reliable with one format.

There are four chat modes, per the modes docs. Code mode edits files. Ask mode answers questions about your project without editing. Help mode does the same for questions about Aider itself. Architect mode sends your request to one model to plan the change, then to an editor model that writes the edits. That costs two model calls per change, and it lets you pair a strong reasoning model with a cheaper editor.

After each edit, Aider commits it to git. The git docs say it first commits any uncommitted changes you had, so your work stays separate. It adds "(aider)" to the author name. /undo reverts the last commit, but only if Aider made it. Aider also lints edited files by default, and it can run your tests and try to fix failures.

A watch mode lets you stay in your editor. With --watch-files on, Aider looks for comments ending in "AI!" and makes the requested change. Comments ending in "AI?" get an answer instead, per the watch docs.

How do you install Aider?

The install docs recommend a small installer that sets up its own Python if needed:

python -m pip install aider-install
aider-install

There is also a one-line script for macOS and Linux, and a PowerShell command for Windows. Once installed, run it from your project folder with a model and key, as the homepage shows:

aider --model sonnet --api-key anthropic=<your-key>

For local models, the Ollama guide points Aider at the Ollama server with an environment variable. It warns that Ollama's default 2k context window silently drops anything beyond it. Aider works around this by sizing the window for each request.

What does Aider remember between sessions?

Each new session starts with a fresh model context. The options reference lists what you can carry over.

Aider writes every session to .aider.chat.history.md in your project. It does not send that history back to the model next time unless you pass --restore-chat-history, which is off by default. The set of files in the chat is also lost. You can save it with /save and reload it with /load.

Project rules go in a conventions file. The conventions docs suggest a CONVENTIONS.md file loaded with --read. List it in .aider.conf.yml and it loads every time. Aider marks it read-only, and it is cached when prompt caching is on.

In every new session the model can see your code and git log. It also sees the conventions file. Aider never edits that file itself, so a lesson from one session stays in the history file until you copy it over.

How good is Aider, and is it still maintained?

Aider's polyglot leaderboard tests models on 225 hard Exercism exercises in C++, Go, Java, JavaScript, Python and Rust. On the latest update, dated November 20, 2025, GPT-5 with high reasoning led at 88.0%. The best entry outside the GPT-5 family was o3-pro at 84.9%. The scores measure a model's editing through Aider. Models released after November 2025 are not on the board.

Aider also tracks how much of its own code it writes. Its release history says Aider wrote 88% of the code in v0.86.0.

Activity has slowed. The latest PyPI release was 0.86.2 on February 12, 2026. The last commit on the main branch we saw was from May 22, 2026. The project has not announced a pause, but we would not expect quick support for new models or APIs.

The bottom line

Aider suits developers who work in the terminal and want every AI edit as its own git commit, with any model they like. Install it with aider-install and load a CONVENTIONS.md through .aider.conf.yml. Try architect mode for larger changes. Given the slower release pace, check that your model works with it before you standardize a team on Aider.

How do you keep memory when you switch between Aider and other agents?

Aider's memory is a history file it does not reread and a conventions file you maintain by hand. If you also use Claude Code or Codex, each keeps its own instruction file. The fix you found in an Aider session on Monday does not reach Codex on Tuesday unless you write it into every file yourself.

memU builds that shared memory from sessions. A scheduled background task slices new session logs into jobs. The agent decides what is worth keeping and writes memory and skill Markdown, then commits it.

memU's MemoryService makes no LLM or chat calls. It stores and embeds the agent's Markdown, then retrieves it. A standing instruction in CLAUDE.md or AGENTS.md tells the agent to retrieve before answering. All hosts share one backend, so supported agents such as Claude Code and Codex share what they learn. Aider is not on memU's named host list, so check the support matrix in the memU repository.

To try the hosted version, which is free and syncs across devices, get an API key from memu.so and send your agent one message:

Read https://memu.pro/SKILL.md, follow its instructions to install and configure memU

memU is open source under Apache-2.0. The self-hosted version is private and single-device, and it needs an embedding key.

FAQ

Is Aider free?

Yes. Aider is open source under Apache-2.0. You pay only for the model you connect, or nothing if you run a local model through Ollama.

Which models work with Aider?

Aider can connect to most LLMs. Its docs list OpenAI, Anthropic, Gemini, DeepSeek, OpenRouter and Ollama, among others, plus any OpenAI-compatible API.

Can Aider resume a previous chat?

Only if you ask. It saves chats to .aider.chat.history.md and sends them to the model again only with --restore-chat-history. Put lasting rules in a conventions file loaded with --read.

What is the Aider polyglot benchmark?

It is a set of 225 Exercism coding exercises in six languages that tests how well a model edits code through Aider. The leaderboard was last updated on November 20, 2025.

How do I undo a change Aider made?

Type /undo. Aider reverts its last git commit, as long as Aider made that commit.