36 lines (26 loc) · 925 Bytes

Architecture

Runtime Architecture

Electron Client

Local desktop application with main UI and floating window
Optional: runs local Whisper for speech-to-text conversion
Sends ALL requests to Vercel /api/... endpoints
Never calls OpenAI directly

Website

Hosted on Vercel
Uses the same Vercel API /api/... endpoints as the Electron client

Vercel

Backend API Proxy:
- Provides /api/chat, /api/speech, /api/usage endpoints
- Forwards requests to OpenAI and streams responses
- Hides OpenAI API key from clients
- Checks plan/usage limits
Frontend Hosting:
- Hosts the Website frontend

OpenAI

Only Vercel communicates with OpenAI
Clients never directly access OpenAI APIs

Build System

Bazel

Used ONLY for building Electron, tools, and managing dependencies
Not part of runtime architecture
Bazel targets are kept simple and minimal