mrcube3/boss

This repository hosts a project the README calls llablab. It is described as a voice first multimodal trading engine. The idea is that a user speaks a command, the system reads a chart image they provide, multiple language models vote on what to do, and an order is sent to a crypto exchange. The README lays out four steps in the pipeline. First, an audio file is sent to Speechmatics, which transcribes the voice command into text. Second, a chart image is sent to Gemini Vision, which classifies the market regime as bullish, bearish, or range bound. Third, three language models accessed through a service called Featherless act as an ensemble and vote on a signal of buy, sell, or hold. Fourth, the Kraken command line tool is used to place the resulting order. The code is split into a backend and a frontend. The backend is a FastAPI service in Python 3.11 or newer that exposes endpoints for health checks, transcription, chart analysis, and running the full pipeline. The frontend is a React 19 app built with Vite 6, described as having a dark and light mode with a glassy iOS style interface. Running the project means starting the FastAPI server on port 8000 and the Vite dev server on port 3000, which proxies API calls to the backend. The README documents a small set of keyboard shortcuts in the UI, including keys for uploading a voice file, uploading a chart image, running the pipeline, resetting, toggling a help guide, and toggling dark mode. API keys for Speechmatics, Featherless, and Gemini are read from an .env file, and the README notes that services without valid keys fall back to a sandbox simulation rather than failing outright. Finally, the repo includes a Dockerfile. A user can build an image tagged voxregime-oracle and run it with the .env file mounted, exposing port 8501. The README does not go into detail about safety controls, position sizing, or risk limits on the trading side beyond describing the pipeline itself.