How to run the harvester
The harvester is the tool that produces the catalog you are reading. If you want to run it yourself — to reproduce a build, to contribute discovery, or just to understand what it does — you can download the source from Codeberg and compile it locally. This page walks through the three commands you need.
What the harvester is
The harvester is a command-line tool that finds candidate AI indicators, extracts them into a common shape, and submits them to a project-operated intake server for review. It is not a scanner: it does not decide whether AI is running on your machine, and it does not send anything anywhere on its own. Its only output is data for the catalog.
What you need
A supported operating system — Linux, macOS, or Windows on x86_64 or ARM64 — with a Go toolchain version 1.23 or newer and `git` installed. No other dependencies. Compilation produces a single static binary you can move where you like.
Build and run
Three commands from a shell:
git clone [https://codeberg.org/traicer/traicer-harvester.git](https://codeberg.org/traicer/traicer-harvester)cd traicer-harvestermake build-harvesterThe last command produces `build/harvester` — a single binary. Run it with `--help` to see the discovery modes it supports.
What running it does
The harvester reads files, package manifests, and public metadata to spot candidate AI indicators. It is read-only on the host: it does not write outside its own working directory, it does not modify the system registry, and it does not spawn elevated processes. Findings are submitted to the intake server for review; you decide what to submit and when.
Where to go next
For the catalog's data model as it ships in a release bundle, see the catalog structure reference, opens in a new tab on Codeberg. To understand how a submission moves through the intake server, read How your contribution reaches the catalog. To read what each review rule the catalog enforces means, see the review-rule reference.