# AI + Glasses Buildouts — Overview

Three AI agent projects Cameron wants to ship, scoped 2026-06-21.

## Status (2026-06-22, scaffolding)

| Project | Status | Where it lives | Priority |
|---|---|---|---|
| **Persian learning agent** | Scaffold built | `/data/cameron/life/projects/persian/` | 1st (smallest build, fastest feedback) |
| **Khanega translation** | Research only; download blocked | `/data/cameron/life/projects/khanega/` | 2nd (after Persian works) |
| **Finance tracker (Plaid)** | Scaffold built | `/data/cameron/life/projects/finances/` | 3rd (bigger infra) |

## Eventually: separate agents
- Cameron wants these as **separate Claude agents** in his fleet, similar to `life_manager` / `glasses` / `vid_model` etc.
- For now: scaffold the apps; agent split can happen once each works as a standalone tool.

## Shared infrastructure
All three want:
- A small Python backend (Flask or FastAPI) — could share a single VPS or run locally on Puget
- `.env` for API keys (chmod 600)
- A small SQLite store for state (or JSON files for simple cases)
- Auth-aware route to glasses dashboard (existing pattern)

## API keys needed (to budget)
- **Claude API** — already have (Cameron's existing Anthropic key) — covers translation + tutoring + categorization + analysis
- **ElevenLabs** — Persian TTS, ~$5-22/mo subscription
- **Azure Speech Services** (fallback / cheap TTS) — pay-per-char
- **Plaid** — Production tier needed for live US Bank, ~$0.30-1/account/mo
- **Whisper API (OpenAI)** — Khanega transcription, ~$0.006/min audio

## Next concrete steps for Cameron
1. **Sign up for Plaid** at https://dashboard.plaid.com/signup → Get client_id + secret → request Production access (takes a few days)
2. **Sign up for ElevenLabs** at https://elevenlabs.io → get API key → try the Persian voice
3. **Decide where to host** — Puget at home, the TRI box, or a small DigitalOcean droplet ($6/mo)
4. **Decide on auth** — these endpoints shouldn't be public

See each project's `README.md` for detailed setup steps.