Murmur

Murmur is a humming-to-song studio. A user hums a sketch, the system transcribes and polishes it into a melody, generates several vibe-led arrangements, then lets the user refine, preview, save, share, and export the result as audio, a poster image, or an audio-backed video (MP4 when supported, WebM as fallback).

For Judges

If you are reviewing this repo with a code bot judge or a design bot judge, start here:

• Product + design + engineering overview: docs/judges-guide.md
• Runtime architecture: docs/architecture.md
• Canonical documentation index: docs/README.md
• Production release operations: docs/repository-operations.md
• Durable paid-generation contract: docs/music-jobs.md
• Runtime surfaces: docs/runtime-surfaces.md
• Delivery cadence: docs/delivery-cadence.md
• Engineering principles: docs/engineering-principles.md
• Review gates: docs/review-gates.md
• Workflow contract: WORKFLOW.md
• Packaging and release: docs/packaging-and-release.md
• Melody, arrangement, and render pipeline: docs/music-engine.md
• Humming engine v2 direction: docs/humming-engine-v2.md
• Humming research landscape + borrowing plan: docs/humming-research-landscape.md
• Audio-engine borrowing deltas: docs/audio-engine-borrowing-deltas.md
• Audio-system closure, fallback, datasets, and supportability: docs/audio-system-closure.md
• Audio architecture loop: docs/audio-architecture-loop.md
• Audio dataset ingestion: docs/audio-dataset-ingestion.md
• Provider and transcription fallback strategy: docs/provider-strategy.md
• Verification notes: docs/verification.md

Architecture

System Overview

flowchart TB
    classDef browser fill:#F5F1EB,stroke:#D2C9B6,stroke-width:2px,color:#1A1A1A
    classDef api fill:#E3F2FD,stroke:#1565C0,stroke-width:2px,color:#0D47A1
    classDef worker fill:#F3E5F5,stroke:#7B1FA2,stroke-width:2px,color:#4A148C
    classDef infra fill:#E8F5E9,stroke:#2E7D32,stroke-width:2px,color:#1B5E20
    classDef accent fill:#FFF3E0,stroke:#FF5924,stroke-width:2px,color:#BF360C
    classDef user fill:#FFF8E1,stroke:#F57F17,stroke-width:2px,color:#E65100
    classDef anchor fill:transparent,stroke:transparent

    User(["🎤 User"]):::user

    subgraph Browser["🌐 Browser  ·  Next.js 16  ·  React 19"]
        direction LR
        Screens["Screens<br/>Hum · Vibe · Studio · Name · Gallery · Song Detail · Me"]
        ClientPitch["WASM pYIN<br/>Essentia.js · browser fallback"]
        ToneSynth["Tone.js · SimpleSynth<br/>live preview & playback"]
        Store["State<br/>Zustand · localStorage"]
        bBot[ ]:::anchor
    end

    subgraph API["⚡ API Routes  ·  Vercel  ·  Node.js"]
        direction LR
        aTop[ ]:::anchor
        Transcribe["/api/transcribe"]
        MusicGen["/api/music/generate<br/>/api/music/jobs"]
        Strummer["/api/strummer/edit"]
        Songs["/api/songs · CRUD"]
        Auth["/api/auth · NextAuth"]
        BillingAPI["/api/billing · Waffo"]
        Notifs["/api/notifications · Web Push"]
        aBot[ ]:::anchor
    end

    subgraph Workers["🐍 Python Workers"]
        direction LR
        wTop[ ]:::anchor
        AE["Audio Engine  ·  Fly.io<br/>RMVPE → SwiftF0 → pYIN<br/>DeepFilterNet denoise · FastAPI"]
        ME["Music Engine  ·  RunPod Serverless<br/>Magenta RT2 · GPU<br/>scale-to-zero · JAX/CUDA"]
        wBot[ ]:::anchor
    end

    subgraph Infra["🗄️ Infrastructure"]
        direction LR
        iTop[ ]:::anchor
        DB[("PostgreSQL<br/>Drizzle ORM")]
        Storage[("Object Storage<br/>S3 / R2")]
        AI[("AI Gateway<br/>OpenAI-compatible")]
        Waffo[("Waffo Pancake<br/>Billing")]
    end

    bBot --- aTop
    aBot --- wTop
    wBot --- iTop

    User -->|"capture hum"| Screens
    Screens -->|"transcribe audio"| Transcribe
    Screens -->|"select vibe"| MusicGen
    Screens -->|"edit arrangement"| Strummer
    Screens -->|"CRUD songs"| Songs
    Screens -.->|"transient failure"| ClientPitch

    ClientPitch -.->|"degraded notes"| Screens
    ToneSynth -->|"live preview"| Screens
    Screens -->|"persist draft"| Store

    Transcribe -->|"raw audio"| AE
    MusicGen -->|"melody + vibe"| ME
    Strummer -->|"prompt"| AI

    Songs --> DB
    BillingAPI --> Waffo
    Transcribe --> Storage
    MusicGen --> Storage
    Notifs --> DB

    AE -->|"contour + notes"| Transcribe
    ME -->|"WAV clip"| MusicGen

Loading

Hum → Song Pipeline

flowchart LR
    classDef stage fill:#FFF3E0,stroke:#FF5924,stroke-width:2px,color:#BF360C
    classDef transform fill:#F5F1EB,stroke:#D2C9B6,stroke-width:2px,color:#1A1A1A
    classDef output fill:#E8F5E9,stroke:#2E7D32,stroke-width:2px,color:#1B5E20
    classDef module fill:#E3F2FD,stroke:#1565C0,stroke-width:2px,color:#0D47A1

    Start(["🎤 Raw Hum"]):::stage

    Transcribe["Audio Worker<br/>pYIN · RMVPE · SwiftF0"]:::module
    Transcribe -->|"RawNote[]"| Polish

    Polish["Melody Polisher<br/>denoise · pitch correction · tonal inference"]:::module
    Polish -->|CleanMelody| Intent

    Intent["Humming Engine<br/>intent / corrected / musical"]:::module
    Intent -->|selected| Versions

    Versions["Generate Versions<br/>3 per vibe · 2 ensembles · seeded"]:::module
    Versions -->|VibeVersion ×3| Assemble

    Assemble["Assemble Song<br/>chords · bass · drums · bpm"]:::module
    Assemble -->|AssembledSong| Live

    Live["Live Preview<br/>SimpleSynth · Web Audio"]:::output
    Assemble -->|AssembledSong| Save

    Save["Export<br/>MP3 · WAV · poster PNG · share HTML · video"]:::output

    Start --> Transcribe

Loading

Transcription Fallback & Resilience

flowchart TB
    classDef tier1 fill:#E8F5E9,stroke:#2E7D32,stroke-width:2px,color:#1B5E20
    classDef tier2 fill:#F3E5F5,stroke:#7B1FA2,stroke-width:2px,color:#4A148C
    classDef tier3 fill:#E3F2FD,stroke:#1565C0,stroke-width:2px,color:#0D47A1
    classDef tier4 fill:#FFF8E1,stroke:#F57F17,stroke-width:2px,color:#E65100
    classDef obs fill:#E0F7FA,stroke:#006064,stroke-width:1px,color:#004D40

    Capture(["🎤 User captures"]):::tier1

    Capture --> Tier1

    subgraph Tier1["Tier 1 · Server (primary)"]
        RMVPE["RMVPE ONNX<br/>Fly.io · GPU-capable"]:::tier1
        RMVPE -->|"confident"| Result["✅ Full Quality<br/>TranscriptionResult"]
    end

    Capture --> Tier2

    subgraph Tier2["Tier 2 · Server (fallback)"]
        SwiftF0["SwiftF0<br/>Fast contour"]:::tier2
        pYIN["pYIN<br/>Conservative"]
    end

    RMVPE -.->|"weak / model absent"| SwiftF0
    SwiftF0 -->|"normal"| Result
    SwiftF0 -.->|"unstable"| pYIN
    pYIN --> Result

    Capture --> Tier3

    subgraph Tier3["Tier 3 · Browser WASM<br/>transient failure only"]
        Essentia["Essentia.js WASM pYIN<br/>lazy-loaded · ~2.5 MB"]:::tier3
        Essentia -->|"degraded"| Degraded["⚠️ Degraded Quality<br/>+ warning in result"]
        Degraded --> Polisher["Melody Polisher<br/>same server-side pipeline"]
        Polisher -->|"usable"| Result
    end

    Capture --> Tier4

    subgraph Tier4["Tier 4 · Explicit Demo<br/>never auto-triggered"]
        Demo["Fixture melody<br/>user chooses 'try an example'"]:::tier4
        Demo --> Result
    end

    subgraph Observability["Observability across all tiers"]
        Latency["Latency Budgets<br/>P50 / P95 per component"]:::obs
        Stage["Stage Tracking<br/>hum → vibe → studio → save → gallery"]:::obs
        Errors["Error Classification<br/>transient.ts · isTransient()"]:::obs
    end

    Result -.-> Observability

Loading

What We Tried To Make Deliberately

• A creation flow with a clear emotional arc: Hum -> Vibe -> Studio -> Name -> Gallery -> Song detail
• A UI tone that feels editorial and restrained rather than tool-heavy: fewer knobs, stronger hierarchy, more guided choices
• A melody pipeline that treats raw humming as imperfect input: denoise, pitch correction, tonal inference, cadence stabilization
• A “what you hear is what you save” architecture: live preview, saved audio, and export all share the same arrangement logic
• Sharing and export that remain useful outside the editor: public song links, poster PNG, audio download, and audio-backed video

Key Visible Files

• Entry flow shell: src/app/page.tsx
• Capture / transcription handoff: src/components/screens/HumScreen.tsx
• Arrangement editing surface: src/components/screens/StudioScreen.tsx
• Saved song playback + export surface: src/components/screens/SongDetailScreen.tsx
• Real audio+video export: src/modules/export/export-video.ts

Getting Started

Install dependencies with Bun:

bun install

If dependency installation stalls on this machine during sharp setup, use:

SHARP_IGNORE_GLOBAL_LIBVIPS=1 bun install

Start the development server:

bun dev

Open http://localhost:3000.

For a fast "is the local stack alive?" pass once web + worker are running:

bun run smoke:local

That smoke check verifies:

• the web app answers on localhost
• /api/user/balance still returns the expected shape
• /api/transcribe fails gracefully with audio_required instead of 500
• the audio worker /health endpoint is alive

For the slightly stronger local operator loop, use:

bun run verify:local

That bundles the stack smoke check with local markdown-link validation, repository lint, and audio-worker unit coverage.

The release-gating browser journey is separate because it builds and boots the production-shaped app. Install Chromium once, then run it with:

bun run test:e2e:install
bun run test:e2e

It exercises capture recovery through Vibe, Studio, save, Gallery, sharing, and public playback with deterministic service boundaries.

For local persistence, start Postgres first:

bun run db:up
bun run db:migrate

If Docker Desktop is installed but not open yet, bun run db:up will fail until the Docker daemon is running.

bun run db:migrate resolves its target through the shared fail-closed DSN resolver (prefer DATABASE_URL_UNPOOLED / POSTGRES_URL_NON_POOLING, then DATABASE_URL / POSTGRES_URL). Locally it uses the .env DATABASE_URL; with no DSN set it errors loudly rather than silently migrating localhost — see Connection string precedence.

For real audio transcription, run the audio worker separately and point the web app at it:

bun run setup:audio
bun run dev:audio

Equivalent manual steps:

cd workers/audio-engine
python3 -m venv .venv
./.venv/bin/python -m ensurepip --upgrade
./.venv/bin/pip install -r requirements.txt
./.venv/bin/uvicorn main:app --host 127.0.0.1 --port 8001

For local audio acceptance and fallback verification, run:

bun run audit:audio
bun run audit:audio:compare
bun run audit:audio:gate
bun run audit:audio:closure

The gate command enforces the checked-in audio baseline at workers/audio-engine/tools/audio_audit_expectations.json and exits non-zero if the shipped path regresses.

To convert a downloaded public dataset or an internal recording folder into a Murmur audit manifest, use the builder documented in docs/audio-dataset-ingestion.md. The closure command uses a suite config so the synthetic baseline is always required while local public datasets and internal golden sets can remain optional until they exist on disk. Recommended local manifest names live under workers/audio-engine/tools/manifests/.

The unattended corpus now includes:

• capture edge cases: quiet / noisy / clipped;
• familiar hooks: two_tigers_phrase, brightest_star_hook;
• structural stress cases: overheld_middle_phrase, pitch_weak_stable_phrase, urgent_hook_fragment.

The summary also reports repair / reroute counts and median pitch latency so the local loop can catch stability and performance drift, not only note-count failures.

For a faster human-readable snapshot, run:

bun run audit:audio:closure:report

For a single unattended acceptance entrypoint that runs the key app-side audio tests, worker acceptance tests, scaffolds the local audio-eval workspace, seeds the local murmur-golden corpus, refreshes the closure report, and writes a combined operator summary, run:

bun run audit:audio:acceptance

It writes:

• workers/audio-engine/tools/reports/audio-closure.md
• workers/audio-engine/tools/reports/audio-acceptance.md
• workers/audio-engine/tools/reports/audio-acceptance.json

If you also want repo-wide lint and build folded into the same run, use:

bun run audit:audio:acceptance:full

That report uses a bounded operator config:

• full synthetic baseline;
• full local humtrans suite when present;
• a limited vocadito slice for readable turnaround;
• the local murmur-golden suite.

For the most reliable local startup, set:

AUDIO_WORKER_URL=http://localhost:8001
AUDIO_ENGINE_PITCH_PROVIDER=auto
MURMUR_ALLOW_DEV_BILLING_FALLBACK=1
MURMUR_DEV_NOTES_BALANCE=9999

auto tries RMVPE first, then falls back to SwiftF0 and pYIN. For RMVPE-backed local work, set AUDIO_ENGINE_RMVPE_MODEL_PATH to a prepared rmvpe.onnx file, or set AUDIO_ENGINE_RMVPE_ALLOW_DOWNLOAD=1 when you explicitly want the worker to fetch the default model. Without an RMVPE model, auto still keeps the local demo alive through the existing SwiftF0 / pYIN fallback. Advanced RMVPE tuning can adjust AUDIO_ENGINE_RMVPE_CONFIDENCE_THRESHOLD; leave it at 0.03 unless you are comparing saved Melo Lab samples.

The base worker keeps local demos light. To enable server denoise, install the optional PyTorch stack and choose the denoise provider explicitly:

pip install -r requirements-denoise.txt
AUDIO_ENGINE_DENOISE_PROVIDER=deepfilternet uvicorn main:app --reload --port 8001

Then set AUDIO_WORKER_URL=http://localhost:8001 in .env. Without the worker, live recordings return a visible retry/demo error instead of silently using a fixture melody.

For local "why does the generated song not match my hum?" work, use MeLo Lab:

bun run dev:audio
bun run dev:music # optional, only for final worker-output drift checks
bun dev

Open /me/debug/melo-lab?debug=1. The lab is a hidden debug-room surface: the page may be visible, but its test APIs send requests only to loopback workers (MELO_LAB_AUDIO_WORKER_URL, MELO_LAB_MUSIC_WORKER_URL) and never route through billing, RunPod/serverless, or the main Hum -> Vibe product path. The Lab uses the same product auto pitch route and renders its returned melody JSON layers through the same browser piano/voice synth before any music-worker probe. Detector details stay in diagnostics/export metadata rather than becoming a Lab selection surface.

Environment Variables

Copy .env.example to .env:

cp .env.example .env

Variable	Description
OPENAI_API_KEY	API key for the default OpenAI-compatible chat endpoint used by /api/strummer/edit.
OPENAI_BASE_URL	Optional override for an OpenAI-compatible base URL.
AI_GATEWAY_API_KEY	Optional alternative to OPENAI_API_KEY when routing through a custom gateway.
AI_GATEWAY_BASE_URL	Optional base URL for a custom AI gateway.
AUTH_SECRET / NEXTAUTH_SECRET	Required in production when OAuth providers are configured.
AUTH_URL / NEXTAUTH_URL / MURMUR_APP_URL	Public origin used for Auth.js redirects, share URLs, and payment return links.
AUDIO_WORKER_URL	Server-only audio worker base URL used by /api/transcribe.
AUDIO_WORKER_TOKEN	Bearer token for Next.js -> audio worker calls; required for production deployments.
MURMUR_CAPTURE_HUMS	Audio-worker raw hum capture. Defaults off; keep unset/false in production unless an explicit review program is running.
AUDIO_ENGINE_PITCH_PROVIDER	Worker pitch detector provider. auto uses RMVPE first, then SwiftF0 and pYIN fallback.
AUDIO_ENGINE_RMVPE_MODEL_PATH	Optional path to a baked rmvpe.onnx model for the RMVPE provider.
AUDIO_ENGINE_RMVPE_DEVICE	RMVPE ONNX Runtime device hint. Defaults to cpu; GPU workers can set cuda or another supported provider.
AUDIO_ENGINE_RMVPE_ALLOW_DOWNLOAD	Optional local/dev opt-in to let rmvpe-onnx download the default model when no model path exists. Defaults off.
AUDIO_ENGINE_RMVPE_CONFIDENCE_THRESHOLD	RMVPE voiced-frame confidence threshold. Defaults to 0.03; tune only with audio sample review.
AUDIO_ENGINE_DENOISE_PROVIDER	Worker denoise provider. auto uses DeepFilterNet when optional deps are installed; deepfilternet fails loudly if they are missing.
MURMUR_ENABLE_MELO_LAB	Explicit production diagnostic flag for the test-only MeLo Lab APIs. Local development enables them by default; worker URLs still must be loopback.
MELO_LAB_AUDIO_WORKER_URL	Optional loopback audio worker override for /api/test/melo-lab/transcribe. Defaults to http://127.0.0.1:8001.
MELO_LAB_MUSIC_WORKER_URL	Optional loopback music worker override for /api/test/melo-lab/music. Defaults to http://127.0.0.1:8002.
DATABASE_URL	Primary Postgres connection string for Drizzle (runtime + migrations). In production point this at the Neon pooler endpoint (*-pooler.<region>.neon.tech) — see Database connections.
POSTGRES_URL	Vercel Postgres-compatible fallback for DATABASE_URL. Accepted everywhere the runtime and DB scripts resolve a DSN, and by the production env audit — one precedence contract, see Database connections.
DATABASE_URL_UNPOOLED	Direct (non-pooler) endpoint used only for migrations, where it takes precedence over the pooled URLs. Set it (or POSTGRES_URL_NON_POOLING) as the production migration secret; the migration workflow fails closed when no DSN is present.
POSTGRES_URL_NON_POOLING	Vercel/Neon-integration name for the direct endpoint; accepted as an unpooled alias in migration context.
MURMUR_DB_POOL_MAX	Optional per-instance postgres-js pool bound. Defaults to 5; only integers from 1 to 10 are accepted. Tune with production load-test evidence.
MURMUR_RATE_LIMIT_DRIVER	Rate-limit backend. Production defaults to postgres for shared route-level buckets; local/test default to memory. redis remains reserved until an adapter lands.
CRON_SECRET	Shared secret for cron routes; production must use a non-placeholder value.
MURMUR_STORAGE_DRIVER	Storage adapter. Production on Vercel must use s3-compatible; dev defaults to local storage.
MURMUR_STORAGE_S3_*	Bucket, region, access key, and secret for the s3-compatible storage adapter.
WAFFO_MERCHANT_ID / WAFFO_PRIVATE_KEY_BASE64 / WAFFO_TOPUP_PRODUCT_ID	Waffo checkout and billing reconcile credentials required for production top-ups.
ZPAY_PID / ZPAY_KEY	Optional WeChat/Alipay route for CNY checkout. If absent, explicit WeChat checkout is unavailable.
MURMUR_ALLOW_PRODUCTION_ZPAY_WITHOUT_REFUNDS	Explicit production allow flag for ZPay checkout while Murmur lacks a reliable ZPay refund/chargeback webhook. Must be 1/true/yes when ZPAY_PID and ZPAY_KEY are set in production; otherwise WeChat checkout stays closed.
RUNPOD_SERVERLESS_ENDPOINT_ID / RUNPOD_API_KEY	RunPod Serverless music-generation endpoint and bearer key.
NEXT_PUBLIC_MURMUR_DURABLE_MUSIC_JOBS	Preview/canary switch for the recoverable paid-generation API. Defaults off; enable with 1 only after migration 0027 and the job telemetry/cutover checks in docs/music-jobs.md.
MURMUR_AUTH_MODE	Auth runtime mode. Defaults to production-like behavior even on localhost: no session means 401. Set demo or local only for explicit preview fallback work.
NEXT_PUBLIC_MURMUR_AUTH_MODE	Browser-side companion for local header auth. Set to local only with MURMUR_AUTH_MODE=local when intentionally exercising localStorage user headers.
MURMUR_ALLOW_HEADER_AUTH	Local/demo-only legacy switch for x-murmur-user-* identity headers. Ignored in production auth mode.
MURMUR_ALLOW_DEV_BILLING_FALLBACK	Development-only switch. Defaults to enabled in next dev; when enabled, local development bypasses notes spending for hum/save/edit flows. Set to 0 to force real billing even in development.
MURMUR_DEV_NOTES_BALANCE	Development-only display balance returned by /api/user/balance and /api/auth/me when dev billing fallback is enabled. Defaults to 9999.

Notes

• Authentication, notifications, and AI now go through Murmur's local platform adapter under src/lib/platform.
• Identity is session-resolved by resolveRequestAuth() in production and by default on localhost. Guest/header fallback is available only after opting into MURMUR_AUTH_MODE=local or demo; x-murmur-user-* is never a production identity source.
• Real recordings go through server /api/transcribe; the fixture melody is only used when the user explicitly chooses the demo action.
• In local development, billing fallback is enabled by default. Hum, save, and Studio edit flows bypass notes spending, and the UI balance defaults to 9999 unless MURMUR_DEV_NOTES_BALANCE overrides it. This bypass is disabled outside development.
• ZPay's current Murmur integration handles successful CNY payment notifications only. Production deployments with ZPAY_PID / ZPAY_KEY must set MURMUR_ALLOW_PRODUCTION_ZPAY_WITHOUT_REFUNDS=1 to acknowledge the temporary refund/reversal gap; otherwise /api/billing/checkout returns 503 zpay_not_configured for explicit WeChat checkout.
• compose.yaml provides the expected local Postgres at postgresql://postgres:password@localhost:5432/myapp.
• The server notification publisher is currently a stub so local development and demo flows stay usable without external push infrastructure. The client still has a local in-app notification inbox and browser alert opt-in for save / generation events.
• The Strummer edit route expects an OpenAI-compatible chat API.

Database connections

The Drizzle client (src/lib/db/client.ts) opens up to 5 Postgres connections per warm serverless instance. On Vercel, each concurrently warm function instance holds its own pool, so the real backend load is:

peak_serverless_instances × max  <  neon_connection_limit

With max: 5, ~20 concurrent warm instances already reach 100 connections — the Neon Pro ceiling, and far past Neon Free's 20. A traffic spike can then exhaust the limit and surface as connection refused across the whole app.

The durable fix is a connection pooler, not a smaller max: in production, point DATABASE_URL at Neon's pooler endpoint (ep-...-pooler.<region>.aws.neon.tech). Neon's PgBouncer multiplexes many client connections onto a small backend pool, so instance count no longer maps 1:1 to backend connections. This is an environment/ops action on the deployment (swap the host in the DATABASE_URL secret). The production env audit rejects direct Neon hosts such as ep-...<region>.aws.neon.tech and accepts pooled Neon hosts; local and non-Neon Postgres hosts are not affected.

MURMUR_DB_POOL_MAX can override the per-instance bound with an integer from 1 to 10. The default remains 5 because there is no production load-test evidence that 1 or 2 preserves request throughput. Use the override only during measured capacity work, and keep peak_serverless_instances × max within the database plan's client-connection limit even when the Neon pooler is enabled.

Connection string precedence

The runtime client, the migration script, and Drizzle Kit all resolve their DSN through one fail-closed helper (resolveServerDsn in src/lib/db/config.ts), so there is a single contract:

Context	Precedence (first non-empty wins)
Runtime (pooled)	DATABASE_URL → POSTGRES_URL
Migrations (prefer direct)	DATABASE_URL_UNPOOLED → POSTGRES_URL_NON_POOLING → DATABASE_URL → POSTGRES_URL

Migrations prefer the direct (unpooled) endpoint because DDL should not run through PgBouncer. When no DSN is configured the resolver throws a clear error instead of silently connecting to localhost — the migration script no longer targets a dev database when an operator forgets the connection string. The localhost default is only used as a fallback under an explicit local-dev signal: NODE_ENV=development / test, or MURMUR_DB_ALLOW_LOCAL_FALLBACK=1.

Stack

• Next.js App Router
• React
• TypeScript
• Tailwind CSS
• Bun
• OpenAI-compatible AI gateway
• Web Audio / Tone-based render pipeline

Deployment

Production: https://murmur.ptoq.io

Current architecture: Next.js frontend on Vercel; transcription (audio-engine) runs on Fly.io; music generation (Magenta RT2) runs on RunPod Serverless (scale-to-zero GPU); Postgres via Drizzle; billing via Waffo.

Production application releases are owned by GitHub Actions and run as CI -> exact-SHA preflight -> evidence approval -> migration -> convergence verify -> exact-SHA deploy -> immutable smoke -> promote -> alias and audio smoke. Vercel's native Git integration may create Previews but must not independently deploy main to Production. See docs/repository-operations.md for the authoritative release contract.

Worker deployment references:

• docs/DEPLOY_MUSIC_ENGINE.md — Vercel shell and worker configuration
• docs/DEPLOY_MUSIC_ENGINE_GPU.md — RunPod Serverless music-engine deploy

Learn More

• Next.js Documentation