▟▙ ● ●
▟██▙ ▌ ‿ ▐
▟████▙ ▗▟███▙
▗▟██████▙▄▄▄▄▟████▙
▝▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▀▛▘
The financial assistant
that lives in your house.
Turi is a locally-hosted market companion — chat with him about your book, and he answers from models running on your own hardware. No cloud required, no keys leaving the machine, and hard-coded rails between any model and your money.
Ask about money. Get answers with receipts.
Quick questions run on the quick-prompt model; deep runs read quotes, news and your thesis notes before they speak. Mention a CLI job mid-chat — "restart the service" — and the assistant hands it straight to Turi's terminal.
Turi runs the floor.
The terminal mascot is also the chief coordinator: he takes your brief, files a ticket, and walks it through 13 review departments — each agent working in its own Docker sandbox. Nothing merges, nothing trades, without clearing the desks.
→ ticket #6 · Intake → … → Release
sandboxed · PR-only · you approve merges
WebUI · CLI · iPhone · Apple Watch
Widgets, Lock Screen and watch complications render from the same data — and the same turi.js sprite spec feeds every terminal, widget and complication.
Your GPU is the backend.
llama.cpp hosts your GGUF models behind a local, OpenAI-compatible endpoint. Auto-tune finds the fastest stable settings for your hardware; a REST API exposes your portfolio, quotes and runs to your own scripts. On iPhone, chat can even fall back to on-device models (Apple Foundation, Llama 3.2) when you're off-network.
api :8789 · Bearer tokens · tailnet-only
Models propose. Rails decide.
Enforced in code, not prompts: agentic account only, equities only, position caps, daily order limits, live re-quotes, and blocked trades become proposals you approve — from the phone, the watch, or not at all.
Everything a market brain needs.
The same book, on every screen.
