Titanium

Titanium is a local-first Electron desktop harness for small AI coding models. It rebuilds the strongest parts of CodeTitan around a much narrower core: Pi-powered agent execution, a clean Electron bridge, persistent sessions, and a polished React UI.

What it does

• Uses @mariozechner/pi-coding-agent as the execution foundation
• Targets local OpenAI-compatible runtimes first: LM Studio, Ollama, or a custom endpoint
• Streams agent activity into the desktop UI in real time
• Stores sessions and config as JSON under Electron userData for easy debugging
• Keeps the runtime small and predictable for weaker local models

Stack

• Electron
• React + Vite
• TypeScript
• Pi coding agent SDK

Project structure

src/
  main/       Electron main process, stores, Pi runtime
  preload/    Safe renderer bridge
  renderer/   React UI
  shared/     Shared types, channels, provider presets
tests/        Store and provider tests

Development

npm install
npm run dev

Build

npm run build

Test

npm test

Notes

• Pick a workspace folder in the UI before starting serious runs.
• Refresh models after starting LM Studio or Ollama so Titanium can auto-detect the active local model.
• The app keeps the tool loop grounded in the configured workspace and uses Pi's native read, bash, edit, and write tools underneath.