First Phase Design

Requirements

To Do List

• All the code should be deployed in GCP at first with some shell script

Request

• Tell me what I should provide for asking you to implement the To Do list.
• Add AI chatting page to the app with some API key and use local ollama in running in local
• Main page should be "/Users/jay/work/task/know.html"
• Google login is required when running in GCP cloud.
• If it runs in local mode it should have be logged in with some specific password.
• Session time should be like 1 hour or something.
• When a user logs in, the investment summary page should be shown that is initially just show a ETH, BTC, S&P, KOSPI index.

Design by Claude

0. Summary

Phase 1 evolves the v0 PoC into a mode-aware, deployable app: one codebase that runs locally (password login + local Ollama) and on GCP Cloud Run (Google login + cloud LLM). It is fronted by know.html as the landing page, shows an investment-summary dashboard (ETH / BTC / S&P 500 / KOSPI) right after login, adds an AI chat page, and ships with a single deploy.sh for GCP.

1. Run modes — one switch, APP_MODE

APP_MODE = local | cloud (default: auto-detect Cloud Run via the K_SERVICE env var).

2. Auth & session (req: Google in cloud, password in local, ~1h session)

• NextAuth / Auth.js v5, JWT session, session.maxAge = 60 * 60 (1 hour) — env-overridable via SESSION_MAX_AGE.
• cloud → Google provider + email allowlist (already built in v0).
• local → Credentials provider comparing one LOCAL_PASSWORD (hashed compare; no user DB).
• middleware.ts guards every route except the public landing; unauthenticated → /login.
• Files: auth.ts (conditional providers by APP_MODE), middleware.ts, app/login/page.tsx.

3. Main page = know.html (req: main page should be know.html)

• Serve know.html as the public landing at /.
• Approach: copy task/know.html → rabbit/public/know.html; / serves it; add a "Log in" entry that routes to /login → on success redirect to /summary.
• ⚠️ Caveat (Q-A): know.html links to local docs/*.md / file:// paths that won't resolve once deployed. Decide: (a) serve as-is (some links dead in cloud), or (b) strip/adapt links to app routes.

4. Investment summary — default post-login view (req: ETH, BTC, S&P, KOSPI)

• Route /summary = landing after login. Four index cards: ETH, BTC (crypto) + S&P 500, KOSPI (equity indices).
• New app/api/indices/route.ts aggregates two sources:
  • ETH / BTC → CoinGecko (reuse existing /api/prices).
  • S&P 500 (^GSPC) + KOSPI (^KS11) → a market-data provider (Alpha Vantage / Twelve Data / Yahoo) → needs MARKET_API_KEY.
• Each card: latest value + daily change %, with a refresh interval.
• The v0 portfolio dashboard (Dashboard.tsx) becomes a secondary tab, not the entry view.

5. AI chat page (req: chat page; API key in cloud, local Ollama in local)

• Route /chat + streaming app/api/chat/route.ts, behind the auth guard.
• Provider abstraction lib/ai.ts:
  • local → POST Ollama /api/chat (OLLAMA_BASE_URL, OLLAMA_MODEL).
  • cloud → hosted LLM with AI_API_KEY (AI_PROVIDER = openai | anthropic | …).
• ⚠️ Local Ollama is currently broken on this machine (missing llama-server backend) — must be fixed before the local path works.

6. GCP deploy via shell script (To-Do: deploy to GCP with a shell script)

scripts/deploy.sh automates the README §9 run book, idempotent (safe to re-run), reading values from a git-ignored scripts/deploy.env:

1. gcloud config set project, enable run / cloudbuild / artifactregistry / secretmanager APIs.
2. create-or-update Secret Manager secrets (auth-secret, google-id/secret, market-key, ai-key).
3. gcloud run deploy rabbit --source . --region … --no-allow-unauthenticated --set-env-vars APP_MODE=cloud,… --set-secrets ….
4. update AUTH_URL to the deployed domain; print the reminder to add the OAuth redirect URI.

7. Config / env matrix

8. New / changed files

auth.ts (conditional providers + 1h session) · middleware.ts · app/login/page.tsx · public/know.html + / route · app/summary/page.tsx + app/api/indices/route.ts · app/chat/page.tsx + app/api/chat/route.ts + lib/ai.ts · lib/market.ts · scripts/deploy.sh + scripts/deploy.env.example · .env.example updates.

9. What you (jay) need to provide (req: "Tell me what I should provide")

Before I implement, please provide / confirm:

1. GCP — project ID, billing enabled, region (default asia-northeast3), gcloud auth login done.
2. Google OAuth — client ID + secret (or OK for me to walk you through creating them), allowed email(s) [linked0@gmail.com?], redirect URIs.
3. Local password — pick LOCAL_PASSWORD (or I generate one).
4. AI — cloud provider choice (OpenAI / Anthropic / …) + API key; local Ollama model name; fix Ollama first.
5. Market data (S&P / KOSPI) — provider choice (Alpha Vantage free tier / Twelve Data) + API key.
6. know.html — serve as-is, or adapt its local links? (Q-A)
7. (optional) custom domain.

AUTH_SECRET I can generate (npx auth secret).

10. Open questions

• Q-A — know.html local-file links: serve as-is (dead links in cloud) or adapt to app routes?
• Q-B — cloud AI provider: OpenAI vs Anthropic vs Gemini?
• Q-C — S&P / KOSPI data source (mind free-tier rate limits)?
• Q-D — is "main page = know.html" the pre-login landing (assumed), with post-login → /summary? Confirm.
• Q-E — local password: single shared value in env (assumed), hashed?

11. Suggested build order

1. APP_MODE switch + auth (Google/cloud, password/local, 1h session).
2. know.html landing + route guard.
3. /summary with 4 indices (+ /api/indices).
4. /chat (Ollama local / API cloud).
5. scripts/deploy.sh + first GCP deploy.