An experiment, and one person built it

Enliner is a working prototype, not a finished product from a large company. One person designed and built it, mostly by directing AI. It runs, and it is measured on real hardware, and there are gaps and rough edges that a bigger team and more time would smooth out. If you buy Enliner, it is a signal to me that you believe in the same things, and it lets me put more time into it alongside other projects.

On-device models have real limits

When you ask a big cloud service to transcribe a call or summarize a document, it reaches for racks of GPUs with terabytes of memory. Enliner does the same job on one computer, on roughly one to two orders of magnitude less compute. For summarizing your own calls and notes, the on-device models are good, and they keep getting better. For the very hardest tasks, the frontier cloud models are still ahead. That gap is real. It will narrow as we keep investing in open model development.

The trade is worth it, and it is ours to make

Winning every benchmark was never the goal. For the things we do every day, keeping our thinking on our own machine is worth more than the last increment of quality. When a job genuinely needs the cloud, use the cloud, on purpose (and having your vault in Markdown makes that easy). The difference is that we make the choice deliberately, rather than having it made for us by someone else.

Some work belongs in the cloud. Most of this does not.

There is real cloud-scale work. Synthesizing a whole vault against a model that needs terabytes of memory and a million-token window is a data-center job, and Enliner will not stop you from using cloud tools for it. Transcribing a phone call is not that. Taking a voice memo is not that. Writing and organizing your notes is not that. A surprising amount of what we have been taught to rent runs perfectly well on hardware we already own.

Built with AI, and that's the point

Enliner's code was written largely by describing what I wanted to AI agents and supervising the work, and making all the decisions. That is the same shift the whole project is about. The tools got cheap enough that one person can build a real, native app, so I used them to build one that then runs entirely on your machine. Using good tools to make something you can own is exactly the point.

I am not against the cloud, and I pay for the best models myself to enable experiments like Enliner, because for some work they earn it. The argument here is narrower and more practical than a rejection of any of it. It is the worst it will ever be, and everything gets better from here: better frontier tools we can use to build our own everyday tools, and to compute for ourselves.

Read the philosophy → Get early access