Last updated: 2026-10-03
Aider is an open-source terminal pair programmer, first committed on 4 April 2023, that edits files in a local git repository through any LLM you connect. It commits every change with a descriptive message, builds a repository map for context, and runs under the Apache-2.0 licence.
About Aider
Aider is a command-line AI pair programmer. Paul Gauthier made the first commit on 4 April 2023, and the code is now published by Aider AI LLC under the Apache-2.0 licence. On 3 October 2026 the GitHub repository showed 49,342 stars and about 5,000 forks, and Gauthier is still the top committer with more than 12,600 contributions.
You start it inside a git repository and name the files you want edited. Aider sends the model those files plus a repository map, a ranked outline of the classes and functions elsewhere in the project, then applies the model's answer as an edit and commits it. A cheaper secondary model writes the commit message, so the history stays readable, and a bad edit is reverted with /undo. Architect mode splits the job in two: one model proposes the change and a second turns it into file edits, which lets you pair a reasoning model from Anthropic or OpenAI with a cheaper editor.
Aider suits developers who already live in a terminal, work over SSH, or want AI changes to arrive as ordinary commits a team can review. It is a weaker fit if you want inline autocomplete, which is where GitHub Copilot is built to help, or an AI-first editor such as Cursor, while Cline offers a similar bring-your-own-key model inside VS Code and Claude Code runs longer autonomous loops in the terminal. Our ranking of AI coding assistants shows how the group compares by job.
It runs on macOS, Linux and Windows through a pip package or the project's one-line uv installers. Local models served by Ollama or LM Studio work too, which keeps code on your own machine when the model is local (see our guide to AI models you can run locally for coding). The directory lists it under code and development tools and in the open-source architecture facet.
Model support is the part that moves fastest. The release notes list Claude 4.5 and 4.6, GPT-5.1 to 5.4 and Gemini 3 preview models from Google under a section headed main branch, with no versioned heading above v0.86.1, so confirm that the version you install knows the model you plan to use. HokAI has profiles for Claude Sonnet 4.6 and GPT-5.2, and the same notes add a DeepSeek Reasoner model from DeepSeek. Treat the homepage badges (44K stars, 6.8M installs, 15B tokens a week) as the project's own figures, which lag the live GitHub count.
Screenshots
Pricing
Aider has no subscription, seat price or paid tier, and its open-source licence allows commercial use. You pay your model provider per token, or only for hardware when the model runs locally. 08 for its 225-exercise run (leaderboard entry dated 23 Aug 2025), which is a useful yardstick for a heavy session.
Other zero-subscription options are collected in HokAI's free-tier AI coding tools list and in our free coding assistants guide.
Key Features
- Repository map: Ranks classes and functions across the project with a graph algorithm and sends the most relevant slice to the model within the --map-tokens budget.
- Git-native commits: Commits each edit with a Conventional Commits message written by the weak model, and /undo discards the last change.
- Four chat modes: Code, ask, architect and help modes; architect pairs a proposing model with an editor model that writes the file edits.
- Any model, 15 setup guides: The docs carry setup pages for 15 providers, from OpenAI and Anthropic to local Ollama and LM Studio, including a generic OpenAI-compatible route.
- Lint and test loop: Runs your linter and test command after each edit and hands failures back to the model to fix.
- Voice and images: /voice turns speech into prompts, and /add or /paste puts screenshots into the chat for vision-capable models.
Pros
- Reviewable by default: every edit is its own commit, so a bad change is one /undo away and the log reads like a teammate's work.
- No lock-in on model or editor: switching from a hosted model to a local one is a flag, and the same session works over SSH.
- The project publishes polyglot benchmark runs with cost and well-formed-edit rates per model, which helps you pick an affordable model before spending.
- Open source code and a public issue tracker let you audit what it sends to a provider and fork it if you disagree with a direction.
Cons
- No native MCP client: the open request titled MCP SUPPORT (issue 3314) has 196 thumbs-up, and two pull requests adding it were closed unmerged, so workflows built on MCP tools need a different tool.
- Pip releases lag the repository: the newest PyPI version, 0.86.2, is dated 12 Feb 2026, while support for newer models is listed only under the unreleased main branch.
- No inline autocomplete: edits arrive as whole-turn changes you review afterwards, so it does not replace a completion tool.
- Commits skip pre-commit hooks by default (--git-commit-verify is false), so a repo that relies on hooks for linting or secret scanning gets unchecked AI commits until you change the flag.
Data Handling
- Training-data policy
- Aider sends your code and prompts to whichever LLM API you configure, so that provider's retention and training terms apply. Aider's own analytics are opt-in and exclude code, prompts, chats and keys.
Frequently Asked Questions
What are Aider's pricing plans in 2026?
Aider itself costs nothing, because there is no paid tier or subscription to choose. Your spend is the model API you connect, billed per token by that provider. Aider's own leaderboard prices one GPT-5 high-reasoning pass over 225 coding exercises at $29.08, roughly 13 cents per exercise, as of the entry dated 23 Aug 2025. Pointing it at a local model through Ollama or LM Studio removes the per-token bill.
Is Aider free to use?
Yes, and nothing is held back behind a plan. The limits come from elsewhere: your model provider's rate limits and context window, and the Python requirement of the pip package, which on PyPI is 3.10 up to 3.12. The official installer fetches its own Python 3.12 if yours is newer or missing.
What are the best alternatives to Aider?
[Cline](/hub/agents/cline) is the nearest bring-your-own-key option if you would rather work inside VS Code or JetBrains than a terminal prompt. [Claude Code](/hub/agents/claude-code) suits people who want a terminal agent that runs longer loops on its own. [Cursor](/hub/tools/cursor) fits those who want an AI-first editor with inline completions bundled into one subscription.
Aider or Cline: which should you pick?
Pick Aider when you want each edit as a separate git commit and a workflow that runs the same over SSH as on a laptop. Pick Cline when you want diffs and approvals shown inside your editor and tool use through MCP, which Aider does not ship natively. If neither constraint applies, try both on the same small task, because both let you bring your own key.
What does it take to start using Aider?
Run python -m pip install aider-install, then aider-install, which sets Aider up in its own environment. Change into a git repository and launch it with a key, for example aider --model sonnet --api-key anthropic=<key>. On Windows the project also offers a PowerShell one-liner that installs it through uv. Add the files you want changed with /add, describe the change, and review the commit it produces.
Top Alternatives
- Cline: Pick Cline for diffs, approvals and MCP tools inside VS Code or JetBrains rather than a terminal prompt.
- Claude Code: Pick Claude Code when you would rather hand off a longer terminal task than steer one commit at a time.
- Cursor: Pick Cursor for a full editor with inline completions bundled into a single subscription.
- GitHub Copilot: Pick GitHub Copilot for unlimited inline completions on a paid plan inside the IDE you already use.