MIT licensed

stackmini

a self-hosted AI appliance. one guided CLI, hailports setup, picks the right build and model for your machine and gets you to a working chat. macOS (apple silicon and intel), windows 10/11, linux.

get it on github support the project

see it work

Rendered terminal of the real hailports setup flow, followed by notes/embeddings and grounded-draft data-flow diagrams. Not a live screen recording.

three ways to run it

fully local. llama.cpp, with the build and model chosen for your GPU and RAM. the reference setup uses MiniCPM-V on a base M4 mac mini with 24GB. see dated measurements on the live board.

hybrid. local first. it escalates to a cloud model only for the hard parts and shows you the tokens it saved.

pure API. OpenAI-compatible, Anthropic or Google. keys live only in your env or keychain.

what is in the box

terminal and browser chat, with image input
web search and fetch
a notes brain with embeddings
a grounded drafter that cites your own notes
runbooks, plus a learning loop that picks up your corrections
an idle-work scheduler and a self-healing watchdog
an iMessage bridge on mac
hailports doctor to check the install

it listens on localhost by default and sends no telemetry.

free download

stackmini setup checklist

every install step, mode, model size, port and server flag from the repo docs on one printable page. drop your email and the link shows up right here.

newsletter

get the build log

new setups, numbers and guides when they ship. no spam, unsubscribe any time.