One model writes it. A different one proves it. You approve.
Coding agents tell you they are finished. Often they are not — a missing translation, a test that asserts nothing, a bug that slipped through. The model that wrote the code is the worst possible judge of whether it is right. RemLinkAgent puts a different vendor's model in the reviewer's seat, runs deterministic gates over the result, and shows you the evidence — from your desk or your phone.
Being straight with you: there is no usable release yet. The verification gates work today — this repository is built with them — but the orchestrator that drives two agents is next, not done. Everything runs on your machine, so that machine has to be on. See the roadmap →