Firefox · Desktop + Android

Stream only the audio from YouTube¶

A razor-focused, privacy-first extension that plays YouTube and YouTube Music as pure audio. It fetches a direct audio stream with no login, hijacks the page <video> element, and keeps playing in the background while it blocks ads and telemetry, skips sponsored segments, and adds real YouTube Music power features.

Install for Firefox View on GitHub Explore the architecture

What it does¶

Audio-only, no video decode

A credentialless ANDROID_VR request returns a direct audio URL, and PlayerHandle swaps it into the page <video>. No video frames are decoded, so battery, heat, and bandwidth all drop sharply.

Background & lock-screen play

Playback keeps running when the tab is hidden or the phone is locked, with OS media controls wired through the Media Session API.

Ghost mode: ad & telemetry block

An allowlist webRequest filter drops first-party YouTube telemetry, and a response rewriter strips ad descriptors from InnerTube player / next responses. Everything fails open to native playback.

Private segment skipping

SponsorBlock ranges are fetched by a four-character hash prefix with credentials omitted, so no plaintext video ID ever leaves the browser.

YouTube Music power features

One shared Web Audio graph applies per-track loudness normalization and a five-band equalizer, with opt-in timed lyrics from LRCLIB.

One-tap audio download

An off-by-default in-player control grabs the preferred direct audio format and hands a validated URL and sanitized filename to the downloads API.

Under the hood¶

YouTube Audio runs as four cooperating layers built from a single strict TypeScript source tree (WXT + Preact), shipping as a Firefox Manifest V2 extension. Each feature fails open: if anything breaks, native YouTube keeps working.

flowchart TB
    subgraph UI["UI  &mdash;  Preact + signals"]
        Popup["Popup quick controls"]
        Options["Options page"]
    end

    subgraph BG["Background  &mdash;  persistent MV2"]
        Telemetry["Telemetry allowlist filter"]
        AdFilter["Ad response rewriter"]
        Sponsor["SponsorBlock hash proxy"]
        Remote["Download + LRCLIB proxy"]
    end

    subgraph Content["Isolated content  &mdash;  document_start"]
        Bridge["Nonce-authenticated bridge"]
        Styles["Cosmetic stylesheet"]
    end

    subgraph Page["MAIN world  &mdash;  page context"]
        VR["Credentialless ANDROID_VR fetch"]
        Handle["PlayerHandle: video.src hijack"]
        WebAudio["Web Audio: loudness + EQ"]
    end

    Store[("browser.storage.local")]
    Video[["YouTube &lt;video&gt;"]]

    Popup <--> Store
    Options <--> Store
    Store --> Bridge
    Bridge <-->|postMessage| Handle
    Content --> BG
    VR --> Handle
    Handle --> Video
    WebAudio --> Video

    classDef aqua fill:#22d3b4,stroke:#0f9e86,color:#04120f;
    class VR,Handle aqua;

MAIN world is the only layer that touches page-owned player APIs. It owns the credentialless fetch, the <video>.src hijack, and the Web Audio graph.
Isolated content owns extension storage and the nonce-authenticated bridge, and injects one cosmetic stylesheet for quality-of-life tweaks.
Background owns every privileged API: the telemetry allowlist, the ad response filter, the SponsorBlock hash proxy, and download / lyrics proxies.
UI is a Preact popup and options page over one storage-backed settings model, with quick controls surfaced first for Firefox Android.

Read the full architecture

Explore the docs¶

Architecture

Layer responsibilities, security boundaries, and per-milestone data-flow diagrams for every feature.

System design

Specifications

The "no spec, no code" contracts for milestones M0 through M7, from the foundation to release infrastructure.

SPEC-001 to SPEC-011

Decisions (ADRs)

Why WXT + TypeScript + Preact, and how single-ID AMO-listed distribution, the unlisted beta channel, and AMO preflight are structured.

Architecture decisions

Research

The deep dives behind each feature: streaming internals, ad blocking, ghost mode, ANDROID_VR, mobile support, and more.

Research notes

Project History

Milestone handoffs (M0 to M7) plus live and mobile Firefox verification results.

Handoffs & verification

Agent Instructions

The documentation-driven protocols this AI-enabled repository follows for every change.

Working agreement

Milestone map¶

Milestone	Feature	Specification	Handoff
M0	Extension foundation	SPEC-001	Handoff
M1	Core audio-only playback	SPEC-002	Handoff
M2a	Ghost telemetry blocking	SPEC-003	Handoff
M2b	YouTube ad blocking	SPEC-004	Handoff
M3a	Private segment skipping	SPEC-005	Handoff
M3b	Quality-of-life controls	SPEC-006	Handoff
M4	YouTube Music extras	SPEC-007	Handoff
M5	Audio download	SPEC-008	Handoff
M6	Design polish	SPEC-009	Handoff
M7	Release infrastructure	SPEC-010	Handoff

New here?

Start with the Architecture overview for the big picture, then dive into any Specification for the exact contract behind a feature.