A hand-rolled agent harness for local LLMs.
It runs on your machine and talks to a model on your machine. When a job wants a bigger brain than the one on your desk, you point it at an API instead — same agent, same sessions, same everything else.
It arrives already knowing how to read your things, and it stays small unless you decide otherwise.
That is the whole pitch for now. The rest of it is moving too quickly to write down honestly, which is also why the beta is small.
- Runs
- On hardware you own
- Models
- Local, or any OpenAI-compatible endpoint
- Extends with
- Maki
- Status
- Private beta
The short version
Less on the plate, on purpose.
Local first, not local only
The default is a model running on your own hardware, over your own network, with nothing leaving it. That is a default, not a cage — anything speaking the OpenAI-compatible shape is one setting away, and the choice is per job rather than per install.
Your context, already wired
Retrieval is not a project you take on afterwards. Your folders and your vault are indexed locally and searchable from the first session — the embeddings, the store and the notes all stay on the machine that made them.
Rolled, not welded
Capability arrives as Maki: self-contained rolls that each bring one thing and its tools with it. Add one, ignore the rest. Nothing in the middle is load-bearing for anything else, so the pieces you skip cost you nothing.
Nothing you didn’t order
It ships with a short list of skills and a small bundle, and it is short on purpose. Add what you want; we won’t bloat you. Good sashimi is a few clean things on a plate — no rice, no filler, nothing to hide behind.
Get access
It’s a private beta.
The repo is private and the build goes out by hand. That is mostly because it is still moving quickly and we would rather talk to the first people running it than field issues from strangers. We let folks in a few at a time.
Tell us what you would point it at and what you would run it on. We read every one of these ourselves.