Most of what I have written here this year has been about one problem: coding agents will produce an enormous amount of work, and almost none of the difficulty is in getting them to produce it. The difficulty is knowing which parts you can keep.

aidd is what I built for that on my own machine. It is a local control panel for AI coding work: projects, runs, audits, recipes, and approvals in one place, against repositories that stay on your disk. It is agent-agnostic. It complements an editor rather than replacing one.

The loop is plan, run, audit, review. You give it a project and a piece of work. It plans the work into records you can read and edit before anything executes. It runs an agent against that plan. It audits the result against the repository’s own gates. It records the outcome and the evidence: which gate passed, which did not, and what the run actually touched. Successful coding runs can complete automatically; blocked or unverified work stops for a decision. It is not a manual accept-or-reject gate after every run.

It does not need an aidd account; the coding backends still need their own authentication. Everything aidd ships is in the repository you clone, and there is no hosted service. The license is FSL-1.1-ALv2: available for any use except making aidd itself available to others as a competing commercial product or service, and each release converts to Apache-2.0 on the second anniversary of its own release date. Spernakit, the application template it builds against, is MIT. The .aidd/ contract that projects carry is an open format. Anything can emit it, and nothing about reading it belongs to me.

Installation is from source, which is the only path I support:

git clone https://github.com/NomadicDaddy/aidd.git
cd aidd
bun install
bun run start:web

That needs Bun 1.4.2 or newer. bun run stop:web stops the detached panel. You can also download the source archive from the latest release if you would rather not clone; I install it that way periodically to check that a stranger’s first five minutes still work. Releases carry notes and GitHub’s own source archives. There are no binaries. I wrote about that change in One Commit, No Parents.

If you have an existing repository rather than a new idea, that is the better first thing to try. aidd can adopt a project without editing its code, read what is there, and file what it finds as work you can look at. I have written about what that path looks like, and there is a short explainer on the project page.

The build proofs repository is public and its CI runs where anyone can read them. The posts in this archive include the runs that failed, the audit findings that turned out to be wrong, and the features I deleted after building them. I would rather you checked than took my word.

There is no hosted product, no managed service, and no support promise. It is one person’s tool, used daily on a fleet of applications, and published because it works for me.

Unattended now publishes from its own address. Same archive, same feed, nothing to re-subscribe to.