Skip to main content

Self-hosted · AGPL-3.0 · one engine, 8 front doors

Turn URLs, files and text into media, knowledge and documents.

Give Content a source and say what you want back. It works out what that source can become, plans and runs the job, and delivers named files into your library — from a browser, a terminal, your AI agent or plain HTTP.

Self-host in four lines LatentNoise/content

Rather not install anything? Try the hosted instance early access

Latest release
v0.8.3
Clients, one contract
8
Sites via yt-dlp
1700+
Stars on HomeTube, its best-known client
1.4k

Demo

HomeTube in Content — paste a YouTube URL, choose what you want, follow the job into your library. The same request works from every other client below.

One source, many artifacts

Describe the results. The engine resolves the rest.

No yt-dlp flags, ffmpeg pipelines, transcription glue or LLM orchestration in the client. Each request says what you want; Content works out what is possible for that source and records where every file came from. 

                       ┌─ My Conference.mp4
                       ├─ My Conference - audio.opus
YouTube URL ──→ Content ──┼─ My Conference - subtitles - en.srt
                       ├─ My Conference - transcript.txt
                       ├─ My Conference - summary.md
                       └─ My Conference - summary.pdf

Web page / .md ──→ Content ──┬─ Article.txt
                       ├─ Article - summary.md
                       ├─ Article - translation.md
                       └─ Article.pdf

  • Video

    Stream selection, remux, fast or frame-accurate cutting, SponsorBlock handling.

  • Audio

    Source audio, Opus, MP3 or M4A from media URLs; local files keep their native track.

  • Subtitles

    Manual or automatic tracks, selected by language.

  • Transcript

    From existing subtitles, or from audio with the optional Whisper runner.

  • Summary

    Transcript or readable text through a local Ollama model or an opt-in cloud key.

  • Translation

    Subtitles or transcripts through an LLM; subtitle timings stay aligned.

  • Chapters

    Declared by the source, or derived from a transcript.

  • Thumbnails

    Published artwork or keyframes extracted from the video.

  • Markdown & PDF

    Readable web pages, text files, PDFs' text layer, and PDF output for summaries and transcripts.

Media acquisition works with YouTube and the 1700+ sites supported by yt-dlp. Summaries, translations and derived chapters need a local Ollama or an opt-in cloud key; unavailable runners are reported up front, never discovered halfway through a job.

The clients

One engine, one contract, 8 front doors

Every client speaks the same request, sees the same resolved capabilities and produces the same artifacts. They differ in ergonomics, not in what they can ask for. Pick one or several.

  • Content Studio

    The whole contract in a form: URLs, uploads, server files and inline text.

    In Docker Compose at :8502

    Read more

  • HomeTube

    The simplest YouTube video or playlist → library experience.

    In Docker Compose at :8501

    Read more

  • Browser extension

    Send the Chrome, Brave, Edge or Vivaldi tab you are on to your engine.

    Load unpacked from the release zip

    Read more

  • MCP server

    Give Claude, an IDE or any MCP agent real artifacts instead of talk.

    uvx content-mcp

    Read more

  • CLI

    Terminals, scripts and cron, with an exit code that carries the outcome.

    uv tool install content-cli

    Read more

  • Python SDK

    Typed sync and async client; the one every other client is built on.

    pip install content-sdk

    Read more

  • REST API

    Any language that speaks HTTP. Stable error codes, honest capabilities.

    /api/v1, Swagger at :8010/docs

    Read more

  • Content Console

    Observe and pilot the engine: jobs, runners, storage, configuration.

    In Docker Compose at :8503

    Read more

Screenshots

Content Studio: the general-purpose request builder listing sources, resolved capabilities and outputs
Content Studio renders what the engine answers for your source: a capability the engine gains appears without a UI release, and what cannot be produced is shown with the server's reason.
The Content browser extension popup on a YouTube page, offering outputs and a destination folder
Browser extension: the page you are watching, sent to your engine, with only the outputs that source can produce.
Content Console showing jobs, runners, storage and configuration
Content Console: jobs with their steps, events and logs; runners and their availability; every variable with its effective value. Secrets are never shown.

Quick start

Docker Compose is the only prerequisite

Published images for amd64 and arm64. Nothing to clone or build.

terminal 
mkdir content && cd content
curl -fsSLO https://raw.githubusercontent.com/LatentNoise/content/main/deploy/docker-compose.yml
curl -fsSL -o .env https://raw.githubusercontent.com/LatentNoise/content/main/.env.example
docker compose up -d

Studio at :8502, HomeTube at :8501, Console at :8503, the API and its Swagger at :8010/docs. Full install guide

For agents, terminals and code

Real artifacts, not just talk

Ask Claude to save a lecture's audio into Talks, produce a transcript and a structured summary, and report where every file landed. The agent never needs shell access or yt-dlp syntax.

MCP server

terminal 
# Claude Code
claude mcp add content \
  --env CONTENT_API_URL=http://localhost:8010 \
  -- uvx content-mcp

CLI

terminal 
uv tool install content-cli
export CONTENT_API_URL=http://nas.local:8010

content analyze "https://www.youtube.com/watch?v=…"
content video "https://…" --height 1080 --subs en,fr --watch
content audio "https://…" --format opus --playlist --watch

Python SDK

python 
from content_sdk import ContentClient, outputs

with ContentClient("http://localhost:8010") as client:
    analysis = client.analyze(outputs.url_source("https://www.youtube.com/watch?v=…"))
    job = client.generate(analysis.id, [outputs.audio_output(), outputs.summary_output()])
    job.wait()
    for artifact in job.artifacts:
        print(artifact.display_filename, artifact.delivered_path)

Why build on Content

Principles that hold

  • Declare intent, not tooling

    Requests describe outputs. yt-dlp, ffmpeg, Whisper, LLMs and PDF renderers stay replaceable implementation details.

  • Self-hosted, local-first

    Workstation, NAS or homelab. No account, no telemetry, no mandatory cloud service.

  • One public contract

    Every client converges on the same engine instead of drifting into parallel feature sets.

  • Human files, not pipeline debris

    Every artifact gets a readable name and can land directly in the library you already use.

  • Observable work

    Jobs expose states, ordered events, progress, logs, provenance, cancellation and retry.

  • Honest capabilities

    'Valid but unsupported', 'unavailable on this source' and 'broken' are different answers, given up front.

Where it comes from

HomeTube is the client people already run. Content is the engine it grew into.

Standalone HomeTube — the ad-free video and playlist downloader for Plex, Jellyfin and Emby — has passed 350,000+ package downloads and keeps being maintained. Inside Content, HomeTube is one client among eight: the same paste-a-URL workflow, on an engine that can also produce transcripts, summaries and documents from the same source.

HomeTube in Content, and what changes Standalone HomeTube

Free, open source, yours to run.

AGPL-3.0. Bug reports, ideas and design feedback are welcome through issues.

Install Content Star on GitHub