Qvora is an AI-powered performance creative platform for agencies. It turns any product URL into a full batch of short-form video ads (9:16), then uses real performance signals to generate better-performing variants in the next round. No briefs to write. No editors to brief. No guesswork.
Product URL → AI Brief → Video Variants → Performance Data → Better Variants
- Scrape — Playwright extracts product copy, visuals, and USPs from any URL
- Brief — GPT-4o generates 3–5 creative angles, each with hooks and scripts
- Generate — FAL.AI (Veo 3.1, Kling 3.0, Runway Gen-4.5, Sora 2) produces video variants; ElevenLabs adds voiceover; HeyGen v3 handles avatar lip-sync
- Post-process — Rust + ffmpeg-sys handles watermarking, captions, reframing, and transcode
- Deliver — Mux HLS streaming with signed playback; Cloudflare R2 for zero-egress asset storage
- Learn (V2) — Meta & TikTok signal ingestion feeds back into the next generation cycle
| Role | Access |
|---|---|
| Agency Media Buyer | Full generation + performance dashboard |
| Agency Creative Director | Full generation + brief editing |
| Agency Account Manager | Read-only reviewer |
DTC Brand Manager is Phase 2 — not built in V1.
| Tier | Price | Variants per Angle |
|---|---|---|
| Starter | $99 / mo | Up to 3 |
| Growth | $149 / mo | Up to 10 |
| Agency | $399 / mo | Unlimited |
7-day free trial · No credit card required · Day 8 locks generation · 30-day data retention on cancel
| Phase | Name | Status |
|---|---|---|
| 0 | Foundation & Infrastructure | |
| 1 | Core Data Layer | asset domain (non-superuser role + per-request scoping, empirical isolation test); 4 other domain packages and the prod DB role cutover still pending, see checklist |
| 2 | URL Ingestion & Brief Engine | ✅ Complete |
| 3 | Video Generation Pipeline | |
| 4 | Brand Kit & Export | |
| 5 | Asset Library & Team | |
| 6 | Platform, Billing & Trial | |
| 7 | Polish, Observability & Launch | |
| 8 | Microservices Domain Extraction | ✅ Complete (this is the Phase 8 defined in docs/06-technical — the implementation checklist's Phase 8 is a different, unrelated post-launch NATS/microservices initiative; these two docs use the same number for different things, worth renaming one) |
This table and the previous version disagreed with each other and with the code in multiple places before Apr 15, 2026 — treat any status claim here as current only as of that reconciliation pass, not a permanent guarantee. Full detail, including what was specifically verified vs. assumed: Implementation checklist
| Service | Language | Responsibility |
|---|---|---|
apps/web |
Next.js 15 + tRPC | Frontend + BFF |
services/api |
Go Echo v4 | REST API, business logic domain packages |
services/worker |
Go + asynq v0.26 | Async job processing (video, TTS, lip-sync) |
services/postprocess |
Rust Axum + ffmpeg-sys | CPU-bound video post-processing only |
All business logic lives in typed domain packages — handlers are thin single-line delegates:
services/api/internal/
domain/
identity/ ← workspace, subscription, trial, seats
brief/ ← scrape ingestion, angle/hook generation, inline edits
asset/ ← asset library CRUD, tagging, search
media/ ← Mux lifecycle, R2 uploads, webhooks
signal/ ← ad platform connections, metrics, fatigue, recommendations, oauth
handler/ ← Echo route handlers (delegates only)
middleware/ ← Clerk JWT, rate-limit, tier enforcement
store/ ← pgxpool + sqlc abstraction
util/ ← UUID, timestamp, Postgres helpers
| Concern | Provider | Notes |
|---|---|---|
| Auth + multi-tenancy | Clerk Organizations | One org = one workspace |
| Database | Supabase Postgres + RLS | sqlc codegen, pgx/v5 driver |
| Cache + rate-limiting | Upstash Redis (HTTP) | Never used for queues |
| Job queues | Railway Redis (TCP) | asynq — BLPOP requires persistent TCP |
| Object storage | Cloudflare R2 | Zero egress; presigned PUT uploads |
| Video streaming | Mux | HLS + signed playback + analytics |
| AI models | Vercel AI SDK v6 | generateObject (GPT-4o), streamText (Claude) |
| Video generation | FAL.AI | fal.queue.submit() async — never fal.subscribe() |
| TTS | ElevenLabs | eleven_v3 (quality) / eleven_flash_v2_5 (speed) |
| Avatar lip-sync | HeyGen v3 | V2V only; v4 does not exist |
| Payments | Stripe | Subscriptions + Entitlements API |
| Scraping | Modal + Playwright | Serverless, pay-per-second |
| Secrets | Doppler | dev / stg / prd environments |
| Observability | Sentry + Better Stack + PostHog + Langfuse | |
| Hosting | Vercel (web) · Railway (Go + Rust) | Turborepo CI/CD via GitHub Actions |
- Two Redis instances — Upstash HTTP = cache only. Railway TCP = asynq only. Never swap.
- SSE is not tRPC — Generation progress uses a standalone Next.js Route Handler.
- Tailwind v4 = CSS-only — All tokens in
@theme {}inglobals.css. Notailwind.config.ts. - HeyGen = v3 —
developers.heygen.com. Any "v4" reference in docs is wrong. - FAL.AI = async queue — Always
fal.queue.submit(), neverfal.subscribe(). - Go = I/O bound. Rust = CPU bound. Rust scope is limited to the video postprocessor.
- Agency-first V1 — DTC features are Phase 2.
src/
apps/
web/ → Next.js 15 App Router + tRPC BFF
packages/
ui/ → shadcn/ui component library
types/ → Shared TypeScript types
config/ → Shared ESLint / TS / Tailwind configs
services/
api/ → Go Echo v4 API (domain layer architecture)
worker/ → Go asynq workers (VideoProvider interface)
postprocess/ → Rust Axum + ffmpeg-sys
docs/
01-brand/ → Brand identity
02-product/ → Product definition + overview
03-market/ → Competitive analysis
04-specs/ → Feature spec + user stories
05-design/ → Design system + UI spec + wireframes
06-technical/ → Architecture + API design + schema
07-implementation/ → Phases + checklist + references
wiki/ → Living architecture wiki (canonical source)
supabase/migrations/ → Database migrations (source of truth)
# Install all workspace dependencies
npm install
# Start all services in development
npm run dev
# Type-check
npm run typecheck
# Lint
npm run lint
# Build all packages
npm run buildGo services:
cd src/services/api && go build ./...
cd src/services/worker && go build ./...| Document | Location |
|---|---|
| Product definition | docs/02-product/Qvora_Product-Definition.md |
| Feature specification | docs/04-specs/Qvora_Feature-Spec.md |
| User stories | docs/04-specs/Qvora_User-Stories.md |
| Architecture stack | docs/06-technical/Qvora_Architecture-Stack.md |
| System architecture | docs/06-technical/Qvora_System-Architecture.md |
| Database schema | docs/06-technical/Qvora_Database-Schema.md |
| Implementation phases | docs/07-implementation/Qvora_Implementation-Phases.md |
| Implementation checklist | docs/07-implementation/IMPLEMENTATION_CHECKLIST.md |
| Implementation references | docs/07-implementation/Qvora_Implementation-References.md |
| Living wiki | wiki/ |
| Copilot instructions | .github/copilot-instructions.md |
| Token | Value | Use |
|---|---|---|
| Volt | #7B2FFF |
Primary CTAs, brand accent |
| Convert Green | #00E87A |
Success states, performance wins |
| Signal Red | #FF3D3D |
Fatigue alerts, warnings |
| Data Blue | #2E9CFF |
Analytics, charts |
| Qvora Black | #0A0A0F |
Primary background |
| Qvora White | #F5F5F7 |
Text, surfaces |
Fonts: Clash Display (hero) · Space Grotesk Bold (display) · Inter (UI) · JetBrains Mono (data)
This repository is private and internal. All planning, architecture, and implementation details are confidential. Do not publish externally without explicit approval.
License: Proprietary. All rights reserved.