GlobalBridge AI

Sovereign real-time translation bridge — 100+ languages, live captions, Whisper dictation, and document translation.
Built on QVAC (Tether local AI) + Keet (P2P meetings). Audio and transcripts never leave the device in Sovereign Mode.

Demo

▶ Watch demo on YouTube — sovereign live captions (Whisper + QVAC + tab audio, ~5 min)

Sovereign egress checklist · Install guide (users) · Quick Start · Project site

Why GlobalBridge AI

Cross-language communication today forces a trade-off: either send voice and text to cloud APIs, or give up live translation in meetings and video calls. GlobalBridge AI removes that trade-off by running speech recognition and translation entirely on the user's machine, with optional integration into Keet (end-to-end encrypted P2P calls) and browser tab capture for YouTube, Zoom, Google Meet, and Microsoft Teams.

Use case	Route	Stack
Text + voice dictation	`/`	Local Whisper → QVAC NMT
Two-way conversation	`/conversation`	Whisper dictation + QVAC
Keet P2P meetings	`/meeting`	Tab audio → Whisper → QVAC → personal subtitles
Live captions (YouTube / Zoom / Meet)	`/live`	Tab audio capture → same pipeline
Documents & PDF	`/document`, `/pdf`	QVAC + local processing
Glossary & history	`/glossary`, `/history`	Local SQLite

Tether QVAC Integration

QVAC provides decentralized, on-device AI via @qvac/sdk. GlobalBridge runs a local QVAC bridge (qvac-service, port 8765) for:

Component	Sovereign Mode (default)	Optional cloud
STT	Faster-Whisper (local, GPU/CPU)	Always local
Translation	QVAC Bergamot NMT + local LLM	Together AI fallback
Meeting summary	QVAC local LLM	Together AI
PDF translation	QVAC + PyMuPDF	Together AI

Set in .env:

LOCAL_PROCESSING_ONLY=true
TRANSLATION_PROVIDER=qvac
ALLOW_CLOUD_FALLBACK=false
QVAC_BRIDGE_URL=http://127.0.0.1:8765

Privacy guarantee (Sovereign Mode): No audio, transcript, or document content is sent to third-party APIs.

Pipeline

Microphone / Tab audio (Chrome)
    → WebSocket /api/v1/ws/live  (localhost only)
    → Faster-Whisper STT
    → QVAC translation bridge (localhost:8765)
    → Live subtitles + auto-exported transcript (.txt, .json, .srt)

HTTP dictation endpoint: POST /api/v1/transcribe (PCM int16 @ 16 kHz → Whisper).

Keet Integration

Keet (Holepunch) enables serverless, E2E-encrypted P2P voice/video. GlobalBridge adds a local subtitle layer on top:

User pastes a Keet room invite (keet://…) and opens the call in Keet.
In GlobalBridge /meeting, user selects native language and counterparty language.
User starts the bridge and shares Keet tab audio in Chrome (with “Share tab audio” enabled).
Each participant sees subtitles in their own language (viewer_lang on the backend).
When the session ends, transcript files download automatically.

Chrome extension: Add from Chrome Web Store — live caption overlay on any tab. Requires desktop app running locally. Dev: Load unpacked.

Desktop app (Windows): Tray launcher — setup wizard, start/stop services, open browser. Build installer with npm run desktop:build. See desktop/README.md and docs/DISTRIBUTION.md for GitHub Releases + Chrome Web Store.

Keet handles the call; GlobalBridge handles local STT + QVAC translation. No Keet SDK is required — integration is via standard browser tab audio capture, keeping the stack simple and fully local.

Builder guide: docs/KEET_PEARS_INTEGRATION.md — deep links, WebSocket config, Pear SDK roadmap.

Participant setup link: /meeting?invite=keet%3A%2F%2F…&from=tr&to=en (or use “Copy setup link” in the UI).

Architecture

newtranslate/
├── backend/                 # FastAPI — STT, translation routing, WebSocket live pipeline
│   ├── api/
│   │   ├── websocket.py     # Queued live caption pipeline
│   │   ├── stt.py           # Whisper dictation HTTP API
│   │   ├── translate.py     # QVAC / cloud translation, meetings API
│   │   └── pdf.py
│   └── services/
│       ├── stt.py           # faster-whisper
│       ├── translation.py   # QVAC client routing
│       ├── live_pipeline.py # Async audio queue (non-blocking WS)
│       ├── meeting_export.py# TXT / JSON / SRT export
│       ├── overlay.py       # viewer_lang, caption state
│       └── qvac_client.py
├── frontend/                # Next.js 15 — UI v2 (EN/TR UI locale)
├── qvac-service/            # @qvac/sdk sidecar (port 8765)
├── desktop/                 # Phase 1 tray launcher (setup + services + extension guide)
├── electron/                # Optional floating caption overlay
└── docker-compose.yml       # Sovereign profile

Quick Start

End users (Windows): See docs/USER_INSTALL.md — download desktop app + Chrome extension (2 steps).

Mobile (Android/iOS): Developer preview — mobile/README.md (Expo + QVAC, physical device required).
Developers: continue below.

Requirements

Tool	Version	Notes
Node.js	20+	`node -v`
Python	3.12+	Used for backend + Whisper
Google Chrome	Latest	Required for tab audio (YouTube / Keet / Zoom)
Git	Any	Clone the repo
NVIDIA GPU + CUDA	Optional	Speeds up Whisper STT

On Windows, install Python from python.org and check “Add Python to PATH”.

Option A — Windows desktop app (easiest)

Download GlobalBridge-AI-*-setup.exe from GitHub Releases
Install → first-run setup runs automatically (Python included, ~5–15 min)
Browser opens at http://localhost:3000 when services are ready
Chrome extension → Add to Chrome (desktop app must be running)

Full guide: docs/USER_INSTALL.md

Option B — Developer install (git clone)

1. First-time install

git clone https://github.com/001453/newtranslate.git
cd newtranslate
npm run setup

npm run setup does the following automatically:

Copies .env.example → .env and syncs to backend/.env
Copies frontend/.env.example → frontend/.env.local
Runs npm install in qvac-service/ and frontend/
Creates backend/.venv (if missing) and installs Python deps from requirements.txt
Creates backend/data/ for uploads and SQLite

No API keys required for Sovereign Mode — QVAC runs locally.

2. Run all services

Single terminal (recommended):

npm run dev

Starts QVAC (8765) + API (8000) + Web (3000) in one window. Press Ctrl+C to stop all.

Option B — separate terminal windows:

Windows:

npm run setup          # skip if already done
.\scripts\start-dev.ps1

Linux / macOS:

npm run setup
chmod +x scripts/start-dev.sh
./scripts/start-dev.sh

Option C — run services individually (debugging):

npm run dev:qvac    # QVAC bridge only  → :8765
npm run dev:api     # FastAPI backend only → :8000
npm run dev:web     # Next.js frontend only → :3000

3. Stop services

Method	Command
Single terminal (`npm run dev`)	Ctrl+C
Stale ports / crashed session	`npm run dev:stop`
Windows separate windows	Close each PowerShell window

npm run dev:stop frees ports 8765, 8000, 3000 before the next start.

4. Verify everything is running

Open in your browser:

URL	Expected
http://localhost:3000	GlobalBridge UI
http://127.0.0.1:8000/health	JSON with `"status": "ok"` and `"qvac_available": true`
http://127.0.0.1:8765/health	QVAC bridge health (if exposed)

In the UI, the top bar should show API and QVAC as connected (green).

If QVAC shows offline → wait 30–60 s on first start (model load), or check the QVAC terminal for errors.

5. Configuration

Files are created by npm run setup. Edit locally — never commit these:

File	Purpose
`.env`	Backend: QVAC, Whisper, privacy, security
`backend/.env`	Auto-synced copy of root `.env`
`frontend/.env.local`	`NEXT_PUBLIC_API_URL=http://localhost:8000`

Sovereign Mode (recommended):

LOCAL_PROCESSING_ONLY=true
TRANSLATION_PROVIDER=qvac
ALLOW_CLOUD_FALLBACK=false
QVAC_BRIDGE_URL=http://127.0.0.1:8765

Security (defaults — safe for local use):

API_BIND_HOST=127.0.0.1          # API not reachable from LAN
API_KEY=                         # empty = no auth on localhost
MAX_UPLOAD_BYTES=52428800        # 50 MB PDF upload limit
MAX_BATCH_UPLOADS=10

To expose the API on your LAN, set API_BIND_HOST=0.0.0.0 and a strong API_KEY. See docs/SECURITY.md.

Whisper on CPU (no GPU):

WHISPER_MODEL=medium
WHISPER_DEVICE=cpu

6. Update after `git pull`

git pull
npm run setup    # reinstalls Python + npm deps if requirements changed
npm run dev

7. Troubleshooting

Problem	Fix
`Backend venv missing`	Run `npm run setup`
Port already in use (`EADDRINUSE`)	`npm run dev:stop` then `npm run dev`
UI: “QVAC / API offline”	Check `NEXT_PUBLIC_API_URL=http://localhost:8000` in `frontend/.env.local`
Dictation / mic no signal	Windows mic not muted; pick correct device in sidebar; use Chrome
Live captions silent	Chrome → share tab → enable “Share tab audio”
Translation 503	QVAC not running — check port 8765
`spawn EINVAL` on Windows	Use `npm run dev` (not raw `npm.cmd` in old scripts)

Health check from terminal:

curl http://127.0.0.1:8000/health

Service ports

Service	Port	URL
QVAC bridge	8765	http://127.0.0.1:8765
Backend API	8000	http://127.0.0.1:8000
Frontend UI	3000	http://localhost:3000

All three must run for full translation. API binds to 127.0.0.1 by default (localhost only).

Docker (Sovereign profile)

docker compose --profile sovereign up --build

For local development, npm run dev is simpler.

Usage

Translation & dictation

Open / — type or use Start dictation (local Whisper, not cloud STT).
Pick source/target language; translation runs through QVAC when Sovereign Mode is on.

Keet meeting

Open /meeting — set My language and Other party language.
Paste Keet invite → open in Keet app.
Start bridge → select Keet tab in Chrome → enable Share tab audio.
Stop → downloads .txt, .json, .srt transcript + local meeting summary.

Live captions (YouTube / Zoom / Meet / Teams)

Open /live — same flow as Keet, but optimized for any browser tab with audio.
Use Chrome; check the setup guide on the page.

Performance targets

Metric	Target
STT latency	300–800 ms
Translation latency	500–1200 ms
End-to-end (live)	≤ 1.5–2 s

CPU-only: set WHISPER_MODEL=medium or distil-large-v3 in .env.

Grant & ecosystem alignment

This project is designed for the Tether / QVAC sovereign AI ecosystem and Holepunch Keet P2P communication:

QVAC: All translation and summarization can run via @qvac/sdk with zero cloud egress.
Keet: Adds multilingual accessibility to private P2P calls without centralized servers.
Open source: MIT license, self-hostable, auditable pipeline.

See docs/GRANT.md for the full Tether grant application (copy-paste form text, milestones, $7,500 USD₮ breakdown). Apply at tether.dev.

See docs/ROADMAP.md for planned features, benchmarks, and distribution timeline.

Contributing

Pull requests only — main is protected. See CONTRIBUTING.md.

License

MIT — see LICENSE.

Security

Local sovereign defaults: API binds to 127.0.0.1, optional API_KEY for LAN deploy. See docs/SECURITY.md.

Support the project

If GlobalBridge AI helps you communicate across languages without sending voice to the cloud, consider:

⭐ Star the repo — helps grant reviewers and contributors find the project
🐛 Report issues or request features
💬 Discussions — share your Keet / YouTube setup (enable Discussions in repo settings)
🔀 Pull requests welcome — see CONTRIBUTING.md

Changelog: CHANGELOG.md · Releases: github.com/001453/newtranslate/releases

Name		Name	Last commit message	Last commit date
Latest commit History 59 Commits
.githooks		.githooks
.github		.github
backend		backend
desktop		desktop
docs		docs
electron		electron
extension		extension
frontend		frontend
mobile		mobile
qvac-service		qvac-service
scripts		scripts
shared		shared
.env.example		.env.example
.gitignore		.gitignore
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
docker-compose.yml		docker-compose.yml
package.json		package.json

Folders and files

Latest commit

History

Repository files navigation

GlobalBridge AI

Demo

Why GlobalBridge AI

Tether QVAC Integration

Pipeline

Keet Integration

Architecture

Quick Start

Requirements

Option A — Windows desktop app (easiest)

Option B — Developer install (git clone)

1. First-time install

2. Run all services

3. Stop services

4. Verify everything is running

5. Configuration

6. Update after git pull

7. Troubleshooting

Service ports

Docker (Sovereign profile)

Usage

Translation & dictation

Keet meeting

Live captions (YouTube / Zoom / Meet / Teams)

Performance targets

Grant & ecosystem alignment

Contributing

License

Security

Support the project

About

Topics

Resources

License

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases 5

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

6. Update after `git pull`

Packages