1loop — a self-hosted AI coding agent: Linux, Windows, Android, and your own models

⚡ The gist in 5 seconds - The 1loop project has been published on GitHub as a coding agent that runs on the user's own machine. - Desktop builds for Linux and Windows are announced, along with a headless mode for servers and an Android client. - The user picks the model themselves, and the app's self-updates are signed. ### 🔍 What was found A repository, tpm28/1loop, has appeared in the open — a coding agent that, according to the project description, runs "on your own machine" and works "with the models you choose." The target platforms listed there are desktop builds for Linux and Windows, a headless server (a build without a graphical interface), and Android. The description also specifically mentions signed self-updates — a mechanism where the app updates itself, and the update packages are verified with a digital signature. The mechanics boil down to three claimed properties. First, local execution: the agent runs in the user's environment, not in someone else's cloud. Second, no hard lock-in to a single model: the choice of provider and specific model is left to whoever deploys 1loop. Third, signed auto-updates, meaning the integrity and authenticity of an update is verified on the client side. The headless variant is clearly aimed at server scenarios where there's no desktop, while the Android build targets running the agent from a phone. Neither a license, nor a list of supported models, nor version or build numbers are provided in the repository description — you'll have to dig into the code and releases for the details. ### 💡 Why it matters Most popular coding agents live in the cloud behind a subscription: you hand your code over to an external service and can only work with the models you've been allowed to use. 1loop hits exactly those two pain points — source code privacy and freedom of model choice (all the way down to local inference, if you have the hardware for it). Signed self-updates are no small detail here: auto-updates are precisely the most common vector for supply-chain compromises, so signature verification is a baseline requirement for any tool you trust with access to your repositories. The Android and headless builds extend the use cases to "agent on your phone" and "agent on your home server." How ready the project is for real-world work is a question for the repository's contents, not its description: the claimed platforms and features need to be verified in practice.