I’ve been keeping an eye on the local-first AI assistant space for a while, and most projects in this area feel like duct-taped Python scripts held together by prayers and pip install. So when [Moltis](https://www.moltis.org/) popped up on [Hacker News](https://news.ycombinator.com/item?id=46993587) yesterday and quickly climbed to 118 points, I had to look into it.

Here’s the pitch: one Rust binary, roughly 60MB, containing about 150,000 lines of code. No Node. No Python. No npm install that downloads half the internet. You just grab the binary and run it. That alone is refreshing. The [GitHub repo](https://github.com/moltis-org/moltis) already has over 750 stars, which tells you people are paying attention.

What makes Moltis interesting is how much it packs into that single binary. It routes between multiple LLM providers — OpenAI, local GGUF and MLX models, Hugging Face — so you’re not locked into one ecosystem. It has built-in memory that combines vector search with full-text search, meaning your assistant actually remembers past conversations and can pull up relevant context. It supports sandboxed execution through Docker, Podman, or Apple Containers, so when the LLM wants to run code, it does it safely inside an isolated environment. And you can talk to it through a web UI, Telegram, or a JSON-RPC API.

But the feature that really caught my attention is what they call self-extending skills. Think of it as the assistant writing its own plugins at runtime. If it encounters a task it doesn’t have a skill for, it can create one on the fly, branch a session to test it, and then keep that skill for future use. It’s a bit like watching software evolve in real time, and honestly, it’s the kind of thing that sounds wild until you see it work.

The whole project is MIT licensed, so you can fork it, host it, modify it, whatever you want. Installation is dead simple — there’s a one-liner curl script, a Homebrew tap, Docker images, and you can even build from source with Cargo if you’re into that. They also have [documentation](https://docs.moltis.org/) that’s surprisingly thorough for a project this young.

Fair warning though: the team is upfront that Moltis is still alpha-stage software. You should be careful with tool permissions and not hand it the keys to your entire system just yet. But as a glimpse of where personal AI infrastructure is heading — a single, portable, self-improving binary that runs on your own hardware — this is one of the more compelling projects I’ve come across recently.