DeepQuery

A full-stack AI research workspace. Ask a question, get a web-grounded answer streamed into chat, and keep the thread. For longer jobs, Deep Search queues work on Redis and a background worker so the UI stays responsive.

The live demo is for chatting and auth. Deep Search needs Redis + a worker, so it runs in a local clone.

deepquery.mp4

Features

• Streamed AI chat with Gemini
• Live web context from Tavily
• Google and GitHub sign-in via Supabase
• Conversation history in Postgres
• Deep Search: async research jobs on Redis + BullMQ
• Worker that writes completed answers back into the thread
• Dark mode, markdown answers, and source links

Tech stack

Layer	Tools
Frontend	React, TypeScript, Tailwind, Bun
Backend	Bun, Express, TypeScript
Database	PostgreSQL, Prisma
Auth	Supabase (Google + GitHub)
Models	Google Gemini
Search	Tavily
Queue	Redis, BullMQ

Architecture

Browser
  React UI
  Supabase OAuth
  Polls Deep Search status

API  :3001
  /ask        Tavily + Gemini stream
  /research   Enqueue job (local only)
  Prisma      Users, chats, jobs

Worker
  Pulls BullMQ jobs
  Searches + generates
  Saves the answer

Redis     job queue
Postgres  durable history

Normal chat:

message → /ask → Tavily → Gemini stream → save conversation

Deep Search (local):

message → /research → ResearchJob + BullMQ
worker  → Tavily → Gemini → save result → UI polls status

Live demo vs local

	Live site	Local clone
Google / GitHub login	Yes	Yes
Skip-login test user	Hidden	http://localhost:3000/local
Streamed chat	Yes	Yes
Deep Search	Shows a “clone the repo” note	Full worker pipeline

The skip-login button and /local route only appear on localhost. Production visitors only see Google and GitHub.

Getting started

Prerequisites

• Bun
• Docker Desktop (Postgres + Redis)
• A Supabase project
• Tavily API key
• Gemini API key

1. Clone and install

git clone https://github.com/Khizarshah01/deepquery.git
cd deepquery

cd backend && bun install
cd ../frontend && bun install

2. Start Postgres and Redis

From the repo root:

docker compose up -d

This starts:

• Postgres at postgresql://deepquery:deepquery@127.0.0.1:5462/deepquery
• Redis at redis://127.0.0.1:6382

Those ports avoid clashing with other local Docker apps.

3. Environment

cp backend/.env.example backend/.env
cp frontend/.env.example frontend/.env

backend/.env:

DATABASE_URL="postgresql://deepquery:deepquery@127.0.0.1:5462/deepquery"
REDIS_URL="redis://127.0.0.1:6382"
PUBLIC_SUPABASE_URL="https://your-project.supabase.co"
SECRET_KEY_BASE="your-supabase-service-role-key"
GEMINI_API_KEY="your-gemini-api-key"
TAVILY_API_KEY="your-tavily-api-key"

frontend/.env:

BUN_PUBLIC_SUPABASE_URL="https://your-project.supabase.co"
BUN_PUBLIC_SUPABASE_PUBLISHABLE_KEY="your-supabase-anon-key"
BUN_PUBLIC_BACKEND_URL="http://localhost:3001"

In Supabase → Authentication → URL Configuration, add:

http://localhost:3000
http://localhost:3000/**

plus your live site URLs. Google’s callback stays:

https://YOUR-PROJECT.supabase.co/auth/v1/callback

4. Database

cd backend
bunx prisma generate
bunx prisma db push

5. Run the three processes

# terminal 1
cd backend && bun run start

# terminal 2
cd backend && bun run worker

# terminal 3
cd frontend && bun run dev

Then open http://localhost:3000.

• Sign in with Google/GitHub, or
• Click Skip login / open http://localhost:3000/local to test without OAuth

Keeping Supabase awake

Free-tier Supabase pauses after about 7 days of inactivity. A cheap SELECT 1 is not enough.

DeepQuery follows supabase-pause-prevention:

• /health and /keep-alive insert or delete a row in a keep_alive table
• They also ping the Supabase Auth API
• .github/workflows/keep-alive.yml hits the live /health URL every 10 minutes

After deploy, run bunx prisma db push on the hosted database so the keep_alive table exists.

Project layout

frontend/     React app (Bun)
backend/      Express API + BullMQ worker
prisma/       schema + migrations (under backend/)
docker-compose.yml

License

MIT. See LICENSE.