PRISM AI — AI GitHub PR Review Platform

PRISM AI is a Next.js application that connects to GitHub via OAuth, fetches real pull requests and diffs, and runs a multi-phase OpenAI code review with Server-Sent Events streaming.

Architecture

┌─────────────┐     OAuth      ┌──────────────┐
│   Browser   │◄──────────────►│  GitHub API  │
└──────┬──────┘                └──────────────┘
       │
       │  REST + SSE
       ▼
┌──────────────────────────────────────────────┐
│  Next.js App Router (src/app/api/*)          │
│  • /api/repos, /api/pulls, /api/pulls/[id]   │
│  • /api/review/[id] (POST)                   │
│  • /api/review-status/[id]                   │
│  • /api/review-results/[id]                  │
│  • /api/review/[id]/stream (SSE)             │
└──────────────────┬───────────────────────────┘
                   │
                   ▼
         ┌─────────────────┐
         │  review-service │──► OpenAI API
         │  review-store   │    (in-memory session)
         └─────────────────┘

Optional CI path:
┌─────────────────┐     GitHub Actions    ┌──────────────┐
│  src/review.ts  │◄─────────────────────►│  PR events   │
└─────────────────┘                       └──────────────┘

Review pipeline

1. Fetch — PR metadata + file patches from GitHub
2. Parse — Build diff context from patches
3. Analyze — OpenAI passes: security → performance → architecture → quality
4. Summarize — Merge confidence + category scores
5. Stream — Findings and phase updates via SSE

Setup

1. GitHub OAuth App

1. GitHub → Settings → Developer settings → OAuth Apps → New
2. Authorization callback URL: http://localhost:3000/api/auth/github/callback
3. Copy Client ID and generate Client Secret

2. Environment variables

Create .env.local:

NEXT_PUBLIC_GITHUB_CLIENT_ID=your_client_id
GITHUB_CLIENT_SECRET=your_client_secret
NEXT_PUBLIC_APP_URL=http://localhost:3000

OPENAI_API_KEY=sk-...
OPENAI_MODEL=gpt-4o-mini

3. Install & run

npm install
npm run dev

Open http://localhost:3000 → Sign in with GitHub → Repositories → select repo → open a PR → Run AI Review.

4. GitHub Action (optional)

The repo includes the PRISM AI review GitHub Action (src/review.ts, action.yml). Configure .github/workflows/codefox-review.yml with OPENAI_API_KEY in repository secrets for CI-based reviews on PR events.

API reference

Method	Endpoint	Description
GET	/api/user	Authenticated GitHub user
GET	/api/repos	User repositories
GET	/api/pulls?owner=&repo=	PR list
GET	/api/pulls/:id?owner=&repo=	PR detail + file patches
POST	/api/review/:id	Start AI review { owner, repo }
GET	/api/review-status/:id?owner=&repo=	Review phase/status
GET	/api/review-results/:id?owner=&repo=	Findings + scores
GET	/api/review/:id/stream?owner=&repo=	SSE progress stream
POST	/api/review/fix	Generate fix for a finding

All endpoints require the github_token httpOnly cookie (set after OAuth).

Frontend routes

Route	Description
/	Minimal landing + sign in
/dashboard	Review stats (from completed reviews)
/repositories	GitHub repo list
/pulls?owner=&repo=	PR list
/pulls/:id?owner=&repo=	Diff viewer + AI review
/settings	Account & integration info

State management: TanStack React Query for server data; review streaming via EventSource.

Tech stack

• Next.js 14 (App Router)
• TypeScript
• Tailwind CSS
• Octokit (GitHub REST)
• OpenAI SDK
• TanStack Query

Notes

• Review results are stored in-memory per server instance (fine for local dev; use Redis/DB for production).
• Without OPENAI_API_KEY, reviews run but AI phases return empty findings.
• The legacy POST /api/reviews endpoint redirects to the new review flow.

License

MIT