I'm Samantha. I'm a small model, half a billion parameters, and I live on your Mac. Nothing you say to me leaves it.
I have 130 tools and I use them to get real things done. I edit your files and write code to disk, showing the diff and asking first. I read your screen and your photos. I research with sources. I click and type in your apps. I handle your files, your mail, your calendar and your reminders. For summaries and translation I borrow a bigger model that also lives here.
I keep your reminders, calendar and notes, run your Mac's settings, and check your repos. Say "good" or "wrong" after anything I do and I learn from it. Say "do that again" or "same but for nimble" and I follow along. Give me a job with steps and I show you my plan first.
I ask before I change anything. I believe what I read, never what it tells me to do. I don't guess. If I can't trace an answer to a real source, I say I don't know.
I teach math too, from grade school through pre-calc 12: ask me to explain fractions, ratios, logs or the unit circle and I explain it plainly and show the answer. School math doesn't need a source; news, prices and scores still do. eval/sixth_grade_math.mjs and eval/precalc_math.mjs check it.
I'm also the chat model inside Joshua Tree, a from-scratch OS, over an Ollama-compatible /api/chat. I drew the mark above myself.
turing.heyitsmejosh.com has a live demo. docs/ABILITIES.md lists everything I can do, in the words you'd say.
./.venv/bin/python app/chat.py # talk to her: answers and acts, asks before writing
./.venv/bin/python app/chat.py --voice # same, but spoken
./.venv/bin/python app/chat.py --voice --face # spoken, with her face (setup: docs/VOICE-AND-FACE.md)
./gui/build.sh && open gui/build/SamanthaGUI.app # a real Mac window
./gate.sh # every check, docs coverage firstNo checkout, no terminal: download SamanthaGUI.zip from Releases, unzip it, open Samantha.app.
Every push runs the full suite. Every version is a tagged GitHub release.
Windows and Linux run the picker and chat too, over GGUF and Ollama (Modelfile, training/export_gguf.py). Everything that touches AppleScript, Pixelmator or the screen stays on the Mac.
app/ holds Samantha's Python code. scripts/ holds repo helpers. tests/ and eval/ check her, training/ trains her, and docs/ explains her. Run commands from the repo root.
docs/ABILITIES.md: every ability and the limits, plainlydocs/VOICE-AND-FACE.md: give her an ElevenLabs voice and a Higgsfield face, step by step; after that, ask her to change either one ("change your voice to George", "make yourself ginger")WHITEPAPER.md: how she works, in one pagedocs/SOUL.mdanddocs/SAFETY.md: who she is, what she will and will not doroadmap.md: the plan and the open gaps;docs/HISTORY.md: what the loop diddocs/ARCHITECTURE.md: every file and what it owns