Cast web video to your TV without screen mirroring. Castor extracts streams, converts incompatible formats, and optionally generates subtitles. Run locally or offload media processing to a NAS.

Castor bundles no content or sources and refuses DRM. Use only content you are authorized to access; see Purpose and disclaimer.

_{Run castor cast to browse titles and cast, without leaving the terminal.}

Install on macOS (other platforms) and find your TV:

brew install --cask stupside/tap/castor
castor scan

Save config.yaml in your working directory:

device:
  name: "Living Room TV"   # exact name from `castor scan`
  type: dlna               # or: chromecast, roku

Cast a page:

castor cast player https://example.com/watch/some-video

castor scan found nothing? See Troubleshooting.

castor cast --dry-run ... prints the streams it found instead of casting. Run castor --help for all flags.

Use a native binary on your TV's network, with these tools on PATH:

Encoding uses a working GPU encoder, falling back to software.

• macOS: brew install --cask stupside/tap/castor.
• Linux: download castor_<version>_linux_amd64.tar.gz (or _arm64) from the latest release and put castor on your PATH.
• Windows: download the release ZIP for your architecture and put castor.exe on PATH. Install tools with winget install Gyan.FFmpeg and winget install Google.Chrome. For SmartScreen, choose More info → Run anyway; allow firewall access on private networks for discovery.
• From source: see CONTRIBUTING.md.

Castor reads config.yaml (or --config <path>), then overlays git-ignored config.local.yaml. Environment variables such as CASTOR_CAST__MAX_HEIGHT=720 override both. CLI casting needs device; other keys have defaults. Keep secrets out of git; see SECURITY.md.

Subtitles are burned into served, read-once video (e.g. DLNA), not self-fetching Chromecast/Roku playback. The appropriate default model downloads once to your cache.

cast:
  subtitles: en            # a language code, or auto to detect it; unset for none
whisper:
  # model_path: ""         # ggml-tiny.en for en, multilingual ggml-tiny otherwise

Set whisper.model_path to use a larger model (e.g. multilingual ggml-base.bin). English-only models reject other languages and auto.

Choose the tallest stream within this ceiling and scale known taller video down:

cast:
  max_height: 2160         # default: 1080

Title commands substitute IDs into your source templates, trying proxies in order. Add only sites you are authorized to use:

sources:
  - proxies: ["https://your-source.example"]   # base URLs, tried in order
    templates:
      movie: "/embed/movie/{itemID}"
      episode: "/embed/tv/{itemID}/{season}-{episode}"

Only the interactive browser needs a TMDB key:

tmdb:
  api_key: "<KEY>"

Run on the server with server.token set:

castor media-server   # listens on :8410 (server.listen)

On your computer:

server:
  url: http://my-nas:8410
  token: "<a long random string, the same on both>"

Your computer discovers and controls the TV; the server processes media. Closing the terminal leaves playback running; Ctrl+C stops it. Cast preferences come from the client, the Whisper model from media. Set server.advertise to an address the TV can reach; use HTTPS outside a trusted network.

Run each command in a separate process; API must reach the TV's network:

castor media-server      # :8410
castor scraping-server   # :8412
castor api-server        # :8411

Set server.url: http://localhost:8410 and scraping.url: http://localhost:8412 for API. Configure independent server.token, scraping.token, and api.token; $TOKEN below is the API token.

curl -X POST http://localhost:8411/castor.v1.DeviceService/ListDevices \
  -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' -d '{}'

curl -X POST http://localhost:8411/castor.v1.CastService/Cast \
  -H "Authorization: Bearer $TOKEN" \
  -H 'Content-Type: application/json' \
  -d '{"target": {"deviceId": "<id from ListDevices>"}, "source": {"stream": {"url": "https://example.com/video.m3u8"}}}'

For a page, use "source": {"pages": {"urls": ["https://example.com/watch"]}}. Stop ends a cast; Watch streams feedback (use buf curl or a generated client). CLI clients connect with api.url and api.token.

Page casts go API → scraping → media; direct streams bypass scraping. Media never receives webpages. Without api.url, CLI starts missing services on loopback. See ARCHITECTURE.md for contracts, ownership, and lifecycle.

Castor sideloads a playback channel:

1. Turn on Developer Mode (once, by hand): on the remote press Home x3, Up x2, Right, Left, Right, Left, Right, enable developer mode, and set a web-server password. The device reboots.
2. Put the password in your config (out of git, see Configuration):

   devices:
     roku:
       password: "<dev-web-server-password>"   # first cast only

3. Cast. Castor sideloads its channel automatically; later casts reuse it.

For a published channel, set devices.roku.app_id instead; no developer mode/password needed. Relayed Roku playback runs about 30 s behind.

Discovery cannot cross VLANs/subnets and is blocked on Android/Termux. Set host to skip discovery:

device:
  name: "Living Room TV"   # now just a label
  type: dlna
  host: 192.168.0.3        # the device's LAN IP

For DLNA, host can be a full description URL (e.g. http://192.168.0.3:9197/dmr). On Android/Termux, leave network.interface empty.

The page's video must start without a click. DRM streams are refused.

Force served playback rather than handing the TV a link:

cast:
  delivery: serve   # "auto" (the default) decides per source

Try once with CASTOR_CAST__DELIVERY=serve; it costs bandwidth and CPU. castor --debug explains delivery decisions.

The image includes Chrome, ffmpeg, and ffprobe. --device /dev/dri enables Intel VA-API; otherwise encoding uses software.

docker run --rm --network host ghcr.io/stupside/castor:latest scan

docker run --rm --network host --device /dev/dri \
  -v "$PWD/config.yaml:/config.yaml" \
  -v castor-cache:/root/.cache \
  ghcr.io/stupside/castor:latest \
  cast player https://example.com/watch/some-video

The cache volume preserves Whisper models. On Linux, put CASTOR_SERVER__TOKEN, CASTOR_SCRAPING__TOKEN, and CASTOR_API__TOKEN in git-ignored .env, then run docker compose up -d. API and media use host networking; scraping is published only on loopback. Remove the /dev/dri mapping on hosts without Intel VA-API. A standalone bridged media container needs -p 8410:8410 and server.advertise set to a device-reachable address.

Tags: :latest (stable), :canary (preview), or a pinned :vX.Y.Z.

No bundled video or sources; no DRM decryption or circumvention. You are responsible for lawful use and compliance with site terms. Provided as-is for lawful, personal, and educational use.

See CONTRIBUTING.md, and ARCHITECTURE.md for how castor works inside.