MeetAI

MeetAi- An AI powered meeting platform

Comment

Meet AI

AI-powered meeting platform built with Python (FastAPI) and Jitsi, with optional OpenAI intelligence, Supabase-backed email verification auth, transcripts, summaries, semantic search, and Razorpay billing scaffolding.

What is implemented

  • FastAPI backend with modular routers
  • Supabase Auth-backed email/password verification with local user sync
  • JWT/web cookie fallback auth when Supabase Auth is not configured
  • Sidebar logout (/logout) that clears auth cookies
  • Agent creation (behavior, personality, interview script)
  • Meeting creation and Jitsi room embedding
  • AI interviewer presence in meeting page
  • Voice interview mode:
    • AI asks question by voice
    • user answers by mic
    • primary mode uses browser speech recognition
    • fallback mode records mic chunks and transcribes on server (/transcripts/transcribe)
    • continuous listening with pause-based turn detection
    • interruption support so the assistant can stop speaking when the user starts speaking again
    • AI responds by voice
  • Live transcript auto-save during interview
  • Post-meeting summary generation with in-app background tasks
  • Rolling in-meeting summary refresh (/meetings/{id}/summary/refresh)
  • Embedding-based transcript search
  • Live meeting Q&A from transcript memory (/meetings/{id}/qa)
  • In-meeting transcript filter/search bar (toolbar search)
  • Memory page shows transcript-line counts per meeting and warns when selected meeting has no transcript data
  • Memory Q&A fallback can answer from recent transcript text when embedding rows are unavailable
  • Razorpay webhook scaffolding and plan model
  • LangGraph-Powered Agents: Interview flow is managed by a stateful graph for better follow-ups and script tracking.
  • LangSmith Tracing: Built-in observability for all AI interactions (Chat, Summaries, RAG).
  • React frontend scaffold (frontend/) for API-driven flow

Tech stack

  • Backend: FastAPI, SQLAlchemy, Jinja templates
  • Frontend: React (Vite) + meeting web page template
  • DB: SQLite (default local) or PostgreSQL
  • Video: Jitsi (meet.jit.si)
  • AI & Agents:
    • LangChain & LangGraph (agentic workflows)
    • LangSmith (observability and tracing)
    • OpenAI SDK (chat/embeddings/audio) with support for OpenAI-compatible providers

Environment setup

Copy env and edit values:

cp .env.example .env

Important keys:

  • OPENAI_API_KEY: set this to use OpenAI for live interview intelligence
  • OPENAI_BASE_URL (optional): point the OpenAI SDK at an OpenAI-compatible provider (OpenRouter, Groq, Together, local, etc.)
  • OPENAI_DEFAULT_HEADERS_JSON (optional): JSON object string of headers forwarded to the OpenAI SDK (useful for OpenRouter attribution)
  • OPENAI_CHAT_MAX_TOKENS (optional): caps chat reply length (default 260)
  • OPENAI_TRANSCRIPTION_MODEL (optional): audio transcription model (default whisper-1)
  • DATABASE_URL: default local is SQLite (sqlite:///./ai_meeting_v2.db)
  • AUTO_CREATE_TABLES=true: easiest local mode
  • FRONTEND_ORIGIN=http://localhost:5173: frontend/main-site URL used after email confirmation
  • APP_BASE_URL: backend/base app URL used for local/dev/prod routing and fallback redirects
  • SUPABASE_URL + SUPABASE_ANON_KEY: enable Supabase Auth for signup/login/email verification
  • GOOGLE_CLIENT_ID + GOOGLE_CLIENT_SECRET: enable Google OAuth login
  • GITHUB_CLIENT_ID + GITHUB_CLIENT_SECRET: enable GitHub OAuth login
  • LANGCHAIN_TRACING_V2: set to true to enable LangSmith tracing
  • LANGCHAIN_API_KEY: your LangSmith API key
  • LANGCHAIN_PROJECT: LangSmith project name (default meet-ai)
  • PASSWORD_RESET_EXPIRE_MINUTES (optional): local-auth password reset link lifetime, default 60

Auth mode

Recommended:

  • Set SUPABASE_URL
  • Set SUPABASE_ANON_KEY
  • In Supabase Auth settings, set Site URL to your app URL
  • Add your app URL to Supabase redirect URLs

When Supabase Auth is configured:

  • sign up and login use Supabase Auth
  • verification emails are sent by Supabase
  • password reset emails are sent by Supabase and redirect back to /reset-password
  • after email confirmation, users are redirected to FRONTEND_ORIGIN first, or APP_BASE_URL if FRONTEND_ORIGIN is empty
  • the app still syncs a local users row for meetings, agents, billing, and ownership

When Supabase Auth is not configured:

  • the app falls back to the older local auth flow
  • forgot/reset password works through the app's own SMTP-based email flow
  • email verification is not handled by Supabase, so production signup verification is not recommended in that mode

OpenAI Mode (recommended)

Set in .env:

  • OPENAI_API_KEY=sk-...
  • Optional (OpenAI-compatible providers):
    • OPENAI_BASE_URL=https://openrouter.ai/api/v1
    • OPENAI_DEFAULT_HEADERS_JSON={"HTTP-Referer":"http://localhost:8001","X-Title":"Meet AI"}

When set, these features run on OpenAI:

  • interview Q&A responses
  • summaries
  • embeddings-based search
  • whisper transcription endpoint (/meetings/{id}/transcripts/transcribe)

Note: If you use a third-party OpenAI-compatible provider, chat/summaries usually work. Embeddings and audio transcription depend on the provider.

Mic + transcription notes

  • The meeting page now prefers browser speech recognition first when supported.
  • If browser speech recognition is unavailable or fails, the app falls back to server transcription via MediaRecorder → /meetings/{id}/transcripts/transcribe when OPENAI_API_KEY is set.
  • For best results, use Chrome or Edge and allow microphone permissions.
  • The meeting page keeps listening continuously, waits for a pause before sending your utterance, and can interrupt AI speech when you start speaking again.
  • If you use an OpenAI-compatible provider, audio transcription may be unsupported even if chat works.

Optional fallback mode

If OPENAI_API_KEY is empty, the app still runs with local fallback logic for interview/summaries/search.

Razorpay Billing (Mandatory for subscription flow)

Set in .env:

  • RAZORPAY_KEY_ID=rzp_...
  • RAZORPAY_KEY_SECRET=...
  • RAZORPAY_WEBHOOK_SECRET=...
  • Important: Your Razorpay Webhook URL in the Razorpay dashboard must be set to https://your-app.onrender.com/billing/webhook.

Local run (recommended first)

source venv/bin/activate
pip install -r requirements.txt
python -m uvicorn app.main:app --reload --host 127.0.0.1 --port 8001

Open:

  • http://localhost:8001/health
  • http://localhost:8001/login

Important: set APP_BASE_URL=http://localhost:8001 in .env when running on port 8001, and add http://localhost:8001/login to Supabase Auth redirect URLs if you are testing email verification locally.

Email verification redirect:

  • After verification, users are redirected to FRONTEND_ORIGIN (or APP_BASE_URL) with /?verified=1.
  • If you are not running a separate frontend locally, set FRONTEND_ORIGIN=http://localhost:8001 so verification returns to the same site.

Optional React frontend

cd frontend
cp .env.example .env
npm install
npm run dev

Open: http://localhost:5173

Optional migration flow (Alembic)

PYTHONPATH=. alembic upgrade head

Email verification fields were added in migration 20260309_01, so run migrations if you already have an existing local DB. The local users table is still used even with Supabase Auth, because the app stores ownership and plan data there. Password reset fields were added in migration 20260423_01.

Troubleshooting migrations

If you previously started the app with AUTO_CREATE_TABLES=true on the same database and later enabled Alembic migrations, you may see errors like DuplicateTable: relation "users" already exists.

Fix options:

  • Reset DB (simplest): delete the DB / create a fresh database, then redeploy.
  • Keep existing DB: run PYTHONPATH=. alembic stamp head once (creates/updates alembic_version without running CREATE TABLE), then redeploy.

Render Free note:

  • Render Free does not support web shell access. The default start command uses scripts/render_start.sh, which auto-stamps head for the common DuplicateTable case and continues boot.

Free deploy mode (single service)

This repo is configured for a free single-service deployment (no managed Redis/Postgres/worker required) using render.yaml.

Render deploy options

  • Recommended (Blueprint / Python service): use render.yaml. No Docker setup is required.
  • Docker service: Render will build using Dockerfile. If you created a Docker service earlier, keep Dockerfile in the repo or switch the service to the Blueprint/Python flow.

Quick steps

  1. Push this repo to GitHub.
  2. In Render, create a new Blueprint and select this repo.
  3. Render will create:
    • meetai-web (FastAPI web service)
  4. Set required env vars in Render dashboard:
    • APP_BASE_URL = your Render web URL
    • FRONTEND_ORIGIN = same Render web URL
    • SUPABASE_URL = your Supabase project URL
    • SUPABASE_ANON_KEY = your Supabase anon key
    • OPENAI_API_KEY
    • RAZORPAY_KEY_ID
    • RAZORPAY_KEY_SECRET
    • RAZORPAY_WEBHOOK_SECRET
    • Optional: OAuth keys (Google/GitHub)
  5. Deploy.

Supabase setup for Render:

  1. In Supabase Auth settings, set Site URL to your Render app URL.
  2. Add both https://your-app.onrender.com AND https://your-app.onrender.com/reset-password to Redirect URLs.
  3. Database Configuration:
    • Use the Supabase Connection Pooler (Port 6543) for Render deployments.
    • Set DATABASE_URL in Render to: postgresql://postgres.YOUR_PROJECT_ID:YOUR_PASSWORD@aws-0-YOUR_REGION.pooler.supabase.com:6543/postgres?sslmode=require
    • Important: If your password contains special characters like @, you must URL-encode them (e.g., @ becomes %40).
    • Important: Do not use double quotes " around environment variable values in the Render dashboard.
  4. No SMTP, Redis, or Celery setup is required for the current default deployment path.

Start command used

  • Web: PYTHONPATH=. gunicorn -k uvicorn.workers.UvicornWorker -w 2 -b 0.0.0.0:$PORT app.main:app

You can reuse this from Procfile on other PaaS platforms as well.

How to test interview quickly

  1. Sign up and login at /login
  2. Use Forgot password? on /login if you want to test the recovery flow
  3. Create agent with an interview_script
  4. Start meeting from dashboard
  5. Click Start Interview
  6. Click Enable Camera & Mic
  7. Click Start Mic and speak
  8. Watch mic state text (Listening..., Listening: ..., or Processing...)
  9. Pause briefly and wait for the AI reply
  10. AI asks/replies by voice
  11. Use top toolbar search in meeting page to filter transcript lines
  12. Use sidebar Logout to sign out
  13. Check transcript list (auto-saves both sides)
  14. Click Refresh under Rolling Summary
  15. Ask a question in Ask Meeting Memory

Inviting guests

Each meeting page has Copy Invite Link, which shares a /join/{invite_token} URL.

Notes:

  • On local dev (127.0.0.1), invite links only work for someone who can reach your machine.
  • On Render (HTTPS), the invite link is public and can be opened by anyone with the URL.

Verify OpenAI is active

  1. Set OPENAI_API_KEY in .env
  2. Restart backend
  3. Start interview and ask a follow-up
  4. If responses feel dynamic/contextual, OpenAI mode is active

Performance notes

  • Chat over WebSocket sends only the most recent context (last ~16 messages) to keep responses fast.
  • Transcript embeddings are computed in a background task so saving a transcript line stays fast.

Current limitation

  • The AI icon/tile in Jitsi is simulated via in-page/hidden client behavior.
  • A fully independent media-stream bot participant in Jitsi requires self-hosted Jitsi bot infrastructure.
  • Free single-service mode uses SQLite filesystem storage; data persistence may reset depending on host platform policy.

Built With

  • agent
  • ai
  • backend
  • database
  • fastapi
  • frontend
  • gunicorn
  • jitsi
  • langchain
  • langgraph
  • langsmith
  • openai
  • openai-compatible
  • production
  • razorpay
  • react
  • sqlalchemy
  • supabase
  • uvicorn
  • vite
Share this project:

Updates