Skip to content

GRANT.md

Latest commit

 

History

History
225 lines (154 loc) · 9.09 KB

GRANT.md

File metadata and controls

225 lines (154 loc) · 9.09 KB
Raw

GlobalBridge AI — Tether Grant Application

Apply at: https://tether.dev → Apply for a grant
Category: Applications on Tether’s open stack · QVAC integration · Pears / Keet (P2P)
Program: Tether Developer Grants — local-first AI & payments
Repository: https://github.com/001453/newtranslate
License: MIT

Copy-paste for application form

Project title

GlobalBridge AI — Sovereign Live Translation with QVAC + Keet P2P

Short description (≤ 280 characters)

Open-source app: local Whisper STT + QVAC @qvac/sdk translation + Keet P2P meeting captions. Zero audio egress in Sovereign Mode. Live subtitles, dictation, meeting export (.txt/.json/.srt). MIT, self-hostable.

Long description

GlobalBridge AI is a sovereign real-time translation bridge. Speech is transcribed locally (Faster-Whisper), translated and summarized via Tether QVAC (@qvac/sdk sidecar on port 8765), and shown as live captions — without sending audio or transcripts to the cloud when Sovereign Mode is enabled.

We integrate with Keet (Holepunch P2P): users run encrypted calls in Keet while GlobalBridge captures tab audio locally and renders personalized subtitles in each participant’s native language (viewer_lang). The same pipeline powers YouTube, Zoom, Meet, and Teams via browser tab capture.

Stack: Python FastAPI, Next.js 15, @qvac/sdk, faster-whisper. Working MVP is open source: https://github.com/001453/newtranslate

Grant funds will deliver deeper Pear/Keet integration, VAD-based caption segmentation, a Chrome extension for live captions, benchmark documentation (see docs/BENCHMARKS.md — ~1.9 s live caption p50 on CPU), and a public demo video for the Tether ecosystem.

Grant category (select / describe)

☑ New applications powered by Tether’s open tech stack
☑ Tooling, integrations (QVAC + Pears/Keet)
☑ Documentation & developer onboarding

Requested amount

7,500 USD₮ (milestone-based, see below)

Payment preference

USD₮

Links

Resource URL
Source code https://github.com/001453/newtranslate
Grant application https://github.com/001453/newtranslate/blob/main/docs/GRANT.md
Submit walkthrough https://github.com/001453/newtranslate/blob/main/docs/GRANT_APPLY.md
Demo site https://001453.github.io/newtranslate/
Benchmarks https://github.com/001453/newtranslate/blob/main/docs/BENCHMARKS.md
QVAC https://qvac.tether.io/
Keet https://keet.io/

Team & contact

Field Value
Applicant name Nihat Cetinkaya
Email nihatcetinkaya077@gmail.com
GitHub https://github.com/001453
Location Turkey

Demo video

Public demo: YouTube · docs/DEMO.md

One-line pitch

GlobalBridge AI is an open-source, sovereign real-time translation and live-caption stack that combines Tether QVAC (on-device AI) with Keet (P2P encrypted calls) so multilingual communication never requires sending voice or transcripts to the cloud.

Problem

  1. Privacy — Enterprises, journalists, and activists cannot send meeting audio to Google/Microsoft cloud STT.
  2. Access — P2P tools like Keet have no built-in multilingual subtitles.
  3. Latency & cost — Cloud translation adds 1–3 s delay and recurring API fees per caption line.
  4. Sovereignty — Regulated users need auditable proof that inference stays on-device.

Most AI still runs on remote servers (data leaves the device, latency, exposure). Tether’s QVAC and Pears/Keet stack offers the opposite: local inference and serverless communication. GlobalBridge connects them into one usable product.

Solution (already built — open source MVP)

Layer Technology Where it runs
Speech-to-text Faster-Whisper User machine
Translation & summary QVAC @qvac/sdk localhost:8765
P2P call Keet (Holepunch) Peer-to-peer, E2E
UI Next.js + WebSocket localhost:3000
Meeting export TXT / JSON / SRT Auto-download on session end

Sovereign Mode (LOCAL_PROCESSING_ONLY=true): no third-party egress for audio, transcripts, or documents.

Architecture

Tab audio / Microphone (Chrome)
  → WebSocket /api/v1/ws/live (localhost)
  → Faster-Whisper STT
  → QVAC bridge (@qvac/sdk) — Bergamot NMT + local LLM
  → Live subtitles + transcript export

Keet flow: User opens Keet call → GlobalBridge /meeting shares Keet tab audio → each viewer sees captions in their language via viewer_lang.

QVAC integration (Tether)

  • Sidecar: qvac-service/server.js using @qvac/sdk
  • Backend client: backend/services/qvac_client.py
  • Uses QVAC for live translation, meeting summaries, and document/PDF tasks
  • Cloud APIs (Together AI) are optional and disabled in Sovereign Mode

Alignment with Tether grants program: Local-first AI, no cloud dependency, open modular stack — exactly what the May 2026 grants announcement describes for QVAC.

Keet / Pears integration (Holepunch)

  • /meeting route with Keet invite links (keet://)
  • Personalized bilingual subtitles without a central media server
  • Grant phase will add Pear SDK research, deeper hooks, and published integration guide

What grant funding delivers (milestones)

ID Deliverable Deadline Payout
M1 Public demo & docs — 5–8 min video on DEMO.md; egress checklist; English setup guide (README) +30 days 1,500 USD₮
M2 QVAC hardening — Fix/unify QVAC STT path; benchmark latency (STT + NMT); publish results in docs/BENCHMARKS.md +60 days 2,000 USD₮
M3 Pear / Keet integration v2 — Pear SDK spike; improved meeting bridge; integration doc for Pears builders +75 days 2,000 USD₮
M4 Chrome extension MVP — Live captions overlay for tab audio (QVAC translation); publish on repo extension/ +90 days 2,000 USD₮
Total ~90 days 7,500 USD₮

M1 partially satisfied by existing repo + README; M1 payout on delivery of demo video + docs per Tether review.

Success metrics

Metric Target
Live caption end-to-end latency ≤ 2 s (documented)
STT languages (Whisper) 99+
Translation via QVAC 100+ language pairs (Bergamot packs)
Data egress in Sovereign Mode Zero (documented)
Keet meeting transcript export TXT + JSON + SRT on session end

Why not a pre-defined bounty?

tether.dev bounties list WDK wallet tasks and ANE/CoreML QVAC tasks. GlobalBridge is an application layer combining QVAC + Pears/Keet for accessibility — a fit for “Applications on Tether’s open tech stack”, not WDK modules.

Roadmap after grant

  1. Electron system-wide overlay
  2. VAD-based speech segmentation (cleaner long-form captions)
  3. Mobile LAN subtitle receiver
  4. Optional WDK integration for paid translation sessions (future)

Budget justification

Tether’s program pays $1,500–$10,000 per defined task. At $7,500 over four milestones, funding covers:

  • Engineering (~120–150 h integration + extension work)
  • Documentation and demo production for ecosystem onboarding
  • No ongoing hosting costs — product is fully local/self-hosted

Checklist before you click Apply

After submission — next engineering focus

  1. M3 — Pear / Keet integration v2 (KEET_PEARS_INTEGRATION.md)
  2. Distribution — GitHub Release desktop installer + Chrome Web Store (DISTRIBUTION.md)
  3. Wait for Tether review; respond with repo + demo links if contacted

References