Aider is an open-source AI pair programmer for the terminal that lets language models work directly in your Git repository. As an AI harness, Aider edits files, commits every change automatically and gives the model a repo map as context; through LiteLLM it connects to almost any model. As of 22 November 2026.
What is Aider?
Aider started in 2023 as a project by Paul Gauthier, who per the GitHub API has made by far the most contributions with more than 12,600 commits. The first version appeared on PyPI in June 2023; the Aider-AI/aider repository is licensed under Apache-2.0 and has 50,221 stars per the GitHub API (as of 22 November 2026). Per the installation guide, you install Aider with python -m pip install aider-install and aider-install, which brings its own Python 3.12 if needed; alternatively via uv or pipx. Then you run aider in your project directory.
Aider runs in the terminal, optionally in the browser (--gui), and in any editor via watch mode: with --watch-files, Aider reacts to code comments ending in AI! (change) or AI? (question).
Is Aider still being developed?
Barely. The latest published version is 0.86.2 from 12 February 2026 per PyPI; the latest release object on GitHub is even v0.86.0 from 9 August 2025. The last commit on the main branch dates from 22 May 2026 per the GitHub API, and the repository has been quiet since. Between February and May 2026, the additions were mainly model profiles, for example for GPT-5.5 and Claude Opus 4.7, which per the release history exist only on the main branch. Factor this maintenance status into any new rollout in your organization.
Aider as an AI harness
Aider deliberately implements the building blocks of a harness differently from autonomous agents: instead of freely callable tools and MCP, it offers fixed functions. You add the files to change with /add, the model returns changes in an edit format, and Aider applies them. Per the documentation on chat modes, Aider starts in code mode; ask answers questions without changes, architect lets one model design the solution and a second editor model write the file edits, and help explains Aider itself.
Git integration is the core: per the Git documentation, Aider commits every change with a message in the Conventional Commits style, first commits any existing uncommitted changes separately, and rolls back its own last commit with /undo. The repo map summarizes the repository’s most important classes, functions and signatures and goes to the model with every request. After changes, Aider lints by default and can run tests to fix errors itself. Headless, --message runs a single task and suits scripts.
Models and model access
Aider connects to almost every provider through the LiteLLM package: Anthropic for Claude, OpenAI for GPT, Google for Gemini, DeepSeek, Azure, Bedrock, Vertex AI, OpenRouter and local models via Ollama. The Aider leaderboards show which models edit reliably. For unknown models without a profile, Aider shows warnings but still works.
Custom endpoints are supported per the documentation on OpenAI-compatible APIs: you set OPENAI_API_BASE and OPENAI_API_KEY and start Aider with --model openai/<model-name>. This lets you run Aider in proxy mode through the AI Gateway, which checks permissions and budgets and assigns every token to a cost center.
Enterprise use: data protection and governance
Aider runs locally and sends the files in the chat, the repo map and the history to the chosen model; for commit messages, diffs and chat history go to the weak model. Per the telemetry documentation, analytics are opt-in and anonymous, contain no code and can be turned off permanently with --analytics-disable; without a setting, Aider asks only a random subset of users.
Two defaults are worth knowing. First, per the documentation, Aider skips pre-commit hooks for auto-commits; if you want to enforce secret scanners or formatters, set --git-commit-verify. Second, there is no sandbox: suggested shell commands run with the user’s permissions after confirmation, and per the source code even --yes-always does not confirm them automatically. There is no user or role management; you enforce central policies through distributed configuration files or an upstream gateway. Source code and prompts belong in your record of processing activities.
Running Aider with herdr and Orca
Aider is not currently supported by herdr: the herdr agent list in version 0.9.5 does not name Aider. Orca lists Aider in its own agent overview with automatic setup, without status reporting or hooks. Because Aider commits every change, it fits Orca’s approach of one Git worktree per task well.
How does Aider differ from Claude Code and Codex?
Aider is a pair programmer, while Claude Code and Codex work as autonomous agents with tools, MCP servers and subagents. With Aider, you decide which files the model sees and changes, and every change lands as a Git commit. Aider is model-agnostic, whereas Claude Code and Codex come from the model vendors Anthropic and OpenAI. If you are looking for a model-agnostic, actively maintained agent with MCP, OpenCode is one option.
Where it fits: developer harness and enterprise harness
Aider is a harness for developers who work in the terminal and with Git. For business users without a command line, CompanyGPT is the matching harness, with MCP tools, enterprise knowledge and roles from Entra ID or Keycloak. The AI Gateway provides the shared cost and governance layer; per the product page, its companion measures agent time for Claude Code, Codex, OpenCode, Pi, GitHub Copilot CLI and Gemini CLI, so you capture Aider through proxy mode. Our overview of AI harnesses covers further coding agents and orchestrators.
