Skip to content

About

Self-hosted video meetings with real-time AI speech translation, multilingual captions, chat and optional AI summaries.

Topics

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

Repository files navigation


Open Translate Meet is a self-hosted video meeting application with optional real-time speech translation. It is built with plain PHP, browser JavaScript and LiveKit, and ships with a standalone administrator interface. No proprietary management panel, application WebSocket server or application database is required.

You control the branding, credentials, meeting policies and data retention. AI features and persistent transcripts are disabled by default.


Features

Key features

  • Meetings: video, microphone, screen sharing, device selection, chat and bounded image/video attachments.
  • Access control: separate participant invitations and private host capabilities, server-side moderation, waiting-room admission, room expiration and permanent closure.
  • Speech translation (optional): Gemini Live translation, in-meeting speaking/listening language changes, reconnection status and an original-audio fallback.
  • Chat translation (optional): Gemini or any OpenAI-compatible text endpoint. The original message always stays available.
  • Transcripts and summaries (optional): per-speaker consent, a separate consent for AI summaries, JSON/text export and evidence-linked draft summaries.
  • Booking (optional): requests, approval, cancellation and rescheduling with local availability or Google Calendar, plus configurable SMTP notifications.
  • Administration: room and booking management, provider status and usage counters. Browser-reported audio and operator-rate estimates are clearly distinguished from provider billing.

Architecture

Architecture overview

LiveKit carries the media and the PHP application controls admission and retained state. A configured Gemini project receives the audio of each translating speaker. Text and summary providers receive only the content needed for the requested feature. See the data-flow and privacy guide.


Quick start

Requirements

  • PHP 8.2+ with curl, fileinfo, gd, mbstring, openssl and session
  • Composer 2
  • HTTPS for real browser media
  • LiveKit Cloud or a self-hosted LiveKit server
  • FFmpeg and FFprobe (only for video attachments)

Install

git clone https://github.com/emreocell/Open-Translate-Meet.git
cd Open-Translate-Meet
composer install --no-dev --no-scripts --no-plugins --prefer-dist

export MEET_PRIVATE_DIR=/absolute/path/outside-the-web-root/meeting-private
php bin/init-config.php

Edit the generated private .env and set APP_URL, LIVEKIT_URL, LIVEKIT_API_KEY and LIVEKIT_API_SECRET. APP_SECRET is already generated; do not replace it with a provider key. Use the same MEET_PRIVATE_DIR in PHP-FPM, Apache or the container environment.

Run locally

php bin/set-admin.php
php bin/check-config.php
php -S localhost:8080 -t . deploy/dev-router.php

Open http://localhost:8080/admin/, sign in, create a room and share its participant link. The private host link grants meeting control, so keep it to yourself.

Note

The PHP development server is for local smoke tests only. For real users, use the supplied Apache or Nginx configuration or the Docker setup described in the self-hosting guide. Configure LiveKit's signed webhook to APP_URL/api/livekit-webhook.php. Self-hosted LiveKit must also set room.auto_create: false.


Integrations

Enable only what you need.

Capability Configuration External requirement
Basic meetings LIVEKIT_URL, key, secret LiveKit Cloud account or your own server
Speech translation TRANSLATION_PROVIDER=gemini, explicit GEMINI_MODEL, API key A model with Live translation and ephemeral credentials
Chat translation and summaries TEXT_PROVIDER, TEXT_MODEL, TEXT_API_KEY A text generation API (OpenAI-compatible endpoints supported)
Persistent transcript TRANSCRIPT_ENABLED=true, host enables capture, speakers opt in Live captions
Booking BOOKING_ENABLED=true Local calendar by default; Google OAuth when selected
Notifications SMTP settings and owner email An SMTP account

No model is selected silently. Model IDs, supported languages, quota and availability change, so choose a model supported by your account using the provider's current documentation. Provider quota and billing apply. An unconfigured translation provider leaves normal meetings fully available. This is not an offline translation engine.


Project layout

Path Purpose
api/ Authentication, room state, provider adapters, booking, transcripts, attachments
assets/js/ Browser meeting, translation, chat, admin and accessibility behavior
assets/i18n/ Interface catalogs in 12 languages
admin/ Standalone administrator interface
privacy/, randevu/ Privacy information and the independent booking page
bin/ Local configuration and administrator commands
deploy/ Apache, Nginx, PHP and development-server examples
tests/ Regression tests and isolated protocol integration tests
LICENSES/ Third-party license texts and notices

Verification

npm ci --ignore-scripts
npm run vendor:verify
npm test
php tests/backend_security.php
php tests/portability.php
php tests/root_security_review.php
php tests/bookings.php
php tests/bookings.php google
python3 tests/integration_backend.py
composer audit
npm audit --omit=dev

Node 22+ is development tooling only; the deployed application serves committed browser assets without a build step. CI also checks PHP/JS syntax, dependency manifests, secrets, browser fixtures and the container image.

Important

The isolated integration tests stub provider protocols. They do not prove real microphone, WebRTC, Gemini, Google or SMTP behavior. See the testing guide for live acceptance checks.


Scope and limits

  • Default capacity is eight participants. Raising it changes a deployment ceiling; it does not demonstrate media capacity or provider quota.
  • The file store uses filesystem locks and targets a single application host, not uncoordinated replicas or eventually consistent network filesystems.
  • Anonymous-device bans can be evaded by clearing browser storage or switching devices. Use a password and host admission when identity matters. See the security model.
  • Target browsers are current desktop Chrome, Edge, Firefox and Safari. Device selection, system-audio sharing and autoplay differ between them, and mobile browsers may suspend media in the background.

Documentation

Guide Description
Self-hosting Docker and production installation
Configuration All settings and defaults
Privacy Data flows and retention
Security model Threat model and residual limits
Integration API Signed server-to-server management API
Testing Automated tests and live acceptance checks
Release checklist Preparing a release

Contributing

Contributions are welcome. Please read CONTRIBUTING.md first, and report vulnerabilities privately as described in SECURITY.md.

License

Licensed under the Apache License 2.0, which permits personal and commercial use, modification and redistribution. See NOTICE and THIRD_PARTY_NOTICES.md for attribution and third-party licenses. Company and provider trademarks retain their own rights.


Made by Emre Öcel

About

Self-hosted video meetings with real-time AI speech translation, multilingual captions, chat and optional AI summaries.

Topics

Resources

Contributing

Security policy

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages