Playbooks · Coding tasks
Should you still start a new project with Aider?
Written October 31, 2026
Aider's every-change-is-a-commit workflow is still the reason to pick it. Its last commit was 22 May 2026 and its last packaged release was in February 2026 — decide with that record in front of you, not with a label about whether the project is maintained.
What Aider does well
Aider is a terminal tool for pair programming with a language model in a local git repository. It builds a map of the whole repository to give the model context, works with most popular programming languages, and can connect to almost any model, including local models. Its distinguishing habit is git discipline: Aider automatically commits changes with sensible commit messages, so every change the model makes is a commit you can diff, revert, or cherry-pick with ordinary git tools. The /diff command shows what changed since the last message, /undo reverts the last commit Aider made, and /commit commits outstanding changes. That discipline is the whole case for Aider. If you want an agent whose work arrives as a readable commit history by default, with no extra tooling, Aider still does that job in the way its homepage describes it. It installs as a Python package, with python -m pip install aider-install followed by aider-install, and it runs against the model you name on the command line.
The maintenance record
The record, read from the project's own repository and package page, is this. The repository's latest commit is 5dc9490, dated 22 May 2026, merging a community pull request that added recent Claude model names to Aider's model list. The repository is not archived, it carries an Apache 2.0 licence, and it has close to 50,000 stars — none of which says anything about current activity. The last tagged GitHub release is v0.86.0, from August 2025, and the latest packaged release on PyPI is 0.86.2, from 12 February 2026. That is the version an install pulls today. The gap to be clear about is therefore two gaps: about five months from the last commit to the end of October 2026, and about eight months from the last packaged release. Aider's release-history page lists newer model support — including Claude 4.5 and 4.6 models, Gemini 3 preview models, and GPT-5.3 and 5.4 variants — under a "main branch" heading with no version attached. That work exists in the repository's history, but it has not been packaged into a release a new user can install, and the main branch itself has not moved since May.
What the record costs a new project is concrete, not abstract. Dependency drift: the packaged release pins its dependencies as of February 2026, and new operating-system and Python releases since then are nobody's job to test against. Slow support for new models: the model list in the packaged release predates the model names added on the main branch, so a model released after February may need a manual model name and settings rather than working out of the box. And no one fixing an edge-case bug: an issue filed against 0.86.2 waits on a maintainer response that the commit record does not show arriving. The homepage still names Claude 3.7 Sonnet, DeepSeek R1, and o3-mini-era models as the models Aider works best with, which is how old the model guidance is. None of this means the installed tool stops working. It means the tool you install is the February 2026 tool, and you should budget as if no fix is coming.
What to do about it
- Already using Aider on a project that works. Staying is the easy call. The commit history is already there, the model and settings are already chosen, and nothing in the record breaks a working setup. Pin the version you run, and keep the git history as the safety net it already is.
- Starting a new project and the commit discipline is the deciding factor. Aider is still a defensible pick, with eyes open: install 0.86.2, pick a model the packaged release already knows, run one small real task end to end before committing to it, and accept that a bug you hit is a bug you work around. The discipline is worth that trade only if you will actually read the commits — if you approve changes without reading them, the discipline buys nothing.
- Starting a new project and you need current model support or a maintained dependency chain more than the commit habit. Pick the alternative that meets that need, and apply the same three checks to it before standardising: latest commit date, latest packaged release date, and whether the repository is archived. Any alternative that fails those checks has the same problem under a different name.
- Before a team rollout on any tool, read the repository yourself. Open the commits page, the releases page, and the package page, and note the three dates. A roundup calling a tool actively maintained is a label; the commit record is the evidence, and where the two disagree, the record wins.
What this page does not claim
- No claim is made here about how fast Aider runs with a local model, or how well it performs with one. The homepage says it can connect to local models; speed and quality on a local model depend on the model, the machine, and the task, and no run behind this page measured them.
- No claim is made here about any other tool's maintenance state. The alternative in the guidance above is a check to run, not a recommendation of a named replacement.
Limits
- The maintenance record is a snapshot of the repository and package page as read for this page. A new commit or release after that reading would change it, and the repository itself is the place to look before deciding.
- A dormant commit record is not the same as a broken tool. Aider installs and runs as packaged; the risk described here is drift and unanswered bugs over time, not immediate failure.
Related playbooks
- What version should I be on
- Check for planted instruction files before you start an agent in a folder you did not fill yourself
- DeepSeek, Exa, Brave, Tavily: which web-search API should your agent actually pay for now that search is metered
- Confine Claude Code writes to your project
Sources
- https://github.com/Aider-AI/aider
- https://pypi.org/project/aider-chat/
- https://aider.chat/
- https://Aider.chat/HISTORY.html
Back to: Coding playbooks hub · Also: Have one agent write the code and a second agent check it in the background · Refactor a messy function with an agent, keeping behaviour