Use paths as seen by the server. If Docker mounts a host directory at /media, enter /media as the
library root, not the host path.
To index files without processing them, scan with No flow and leave automatic healthchecks and MKV statistics off. Set original retention before assigning a flow.
Use Dry run on a library to preview its assigned flow, or select files to preview any saved flow. The preview doesn't change media. Deferred means the result depends on processing output or an encoder choice the preview can't determine.
Closing a batch preview doesn't cancel it. Export the CSV before restarting the server if you need the report. Use the flow recipes for examples to try on a few files before processing an entire library.
Watch picks up local filesystem changes but can miss writes made by other machines on a network share. Use recurring scans for those shares, or webhooks for Sonarr and Radarr imports and renames.
Scans ignore hidden files and folders. Check extension and folder filters under Configuration -> Settings if a file doesn't appear.
Use Cooldown to give incoming files time to finish copying. Manual transcodes, manual healthchecks, retries, and Skip cooldown bypass that wait, not pauses, schedules, or node limits. Schedules use the server timezone.
To process only new additions, finish the initial scan with no flow. Set a cutoff after that scan and select Added to Millwright before assigning the flow. Copy and download tools can preserve old modification dates. Cutoffs don't apply to manual requests or jobs already requested.
Pause activity prevents new scans and processing across the instance and survives a restart. Running jobs finish. Cancel a job or scan to stop it. If both the instance and a library are paused, resume both before expecting work from that library.
Healthchecks read and decode the file. A failed tool run isn't a confirmed media error. Check the job log for the cause. If hardware decoding fails or reports errors, the check retries on the CPU. A successful manual check can also release a conversion waiting for that file.
Important
Server-side MKV statistics modify the library file in place. Cancelling the job doesn't undo tags already written. Node-side statistics work on a copy before replacing the original.
Search Jobs by job_id:123 or file_id:123. The file ID also works on Files.
Job snapshots contain metadata, not copies of your media.
Requeuing keeps previous attempts. Jobs started with a selected flow reuse its original settings. Bulk transcodes use each file's library flow. Deleting a job removes history, not media or unused workspaces.
Deleting a library cancels its active work and removes its file list. It keeps the media, completed job history, and retained originals in Trash.
If the browser can't play the original directly, Millwright can convert it for playback without changing the library file. For playback failures, copy playback diagnostics while the problem is happening.
Savings count jobs, so converting one file twice can contribute twice. Deleting job history can change past totals. File-added charts count files still in the library. Language charts count tracks, not files.
Run the CLI from the directory containing compose.yml:
docker compose exec server millwright libraries listUse millwright --help to list commands and millwright --json for scripts.
Set MW_API_URL to your server address and MW_API_TOKEN to the instance API key. The installer
stores it in config/millwright/secrets/api_key. The key grants administrator access. Use HTTPS.
export MW_API_URL=https://millwright.example.org
export MW_API_TOKEN="$(cat /path/to/api_key)"
docker run --rm -e MW_API_URL -e MW_API_TOKEN --entrypoint millwright \
ghcr.io/arabcoders/millwright:latest libraries listHTTP clients send the same key as Authorization: Bearer API_KEY. To browse endpoint examples and
download the OpenAPI schema, enable the API reference.
Use Live API updates to receive progress and status changes in another client.
For failed jobs, startup problems, and logs, see Troubleshooting.