ContextPilot

ContextPilot is an AI-powered Chrome side-panel assistant that understands the current webpage and can summarize, answer questions, draft content, and (optionally) help fill forms using saved profile details.

This repo contains:

• backend/: Node.js + Express API, agent loop + tools, Supabase persistence, Gemini/Anthropic provider wrapper.
• extension/: Chrome Extension (Manifest V3) side panel UI + content extraction + form scanning/fill.

---

Requirements

• Node.js 20+ (recommended: latest LTS)
• A Supabase project (URL + keys)
• A Gemini API key (Google AI Studio / Gemini API)

---

Quick start (local)

1) Backend setup

cd backend
npm install

Create backend/.env (example keys shown; do not commit secrets):

PORT=3001
NODE_ENV=development

AI_PROVIDER=gemini
GEMINI_API_KEY=YOUR_GEMINI_KEY
GEMINI_MODEL=gemini-2.5-flash

SUPABASE_URL=YOUR_SUPABASE_URL
SUPABASE_ANON_KEY=YOUR_SUPABASE_ANON_KEY
SUPABASE_SERVICE_ROLE_KEY=YOUR_SUPABASE_SERVICE_ROLE_KEY

EXTENSION_ORIGIN=*
MAX_PAGE_CONTENT_CHARS=8000

Run the backend:

npm run dev

Health check:

curl -s http://localhost:3001/health

2) Load the Chrome extension

1. Open chrome://extensions
2. Enable Developer mode
3. Click Load unpacked
4. Select the extension/ folder
5. Open the ContextPilot side panel on any webpage

If you deploy the backend (e.g., Railway), update extension/config.js:

• API_BASE_URL: set to your deployed URL

---

How to use

Common prompts

• “Summarize this page”
• “Answer this question using this page: …”
• “Draft a professional reply to this email”
• “Extract pricing/specs from this page”

Form intelligence (optional)

Recommended flow for best reliability:

1. Save details:

“Save my details: Name: …, Email: …, Phone: …”

2. On a form page:

“Fill this form for me”

The assistant will prepare a review payload and the extension will fill fields (it will not submit the form).

---

Testing

Backend test assets live in backend/test/.

API test suite:

cd backend
bash test/test-api.sh http://localhost:3001

Environment check:

cd backend
bash test/check-env.sh

---

Troubleshooting (Gemini)

503 “high demand”

This is Gemini model overload. ContextPilot retries automatically and can fall back to a more stable Gemini model when needed.

429 “RESOURCE_EXHAUSTED”

This is a quota/rate-limit response. ContextPilot respects Gemini’s retry delay and avoids quota-burning retries.

If you consistently see 429:

• Verify your quotas at https://ai.dev/rate-limit
• Reduce request volume (avoid sending multiple prompts rapidly)
• Wait the Retry-After seconds shown and retry

---

Security notes

• Never commit .env files or API keys.
• Use SUPABASE_SERVICE_ROLE_KEY only on the server (backend).