Skip to content

Backups

github-actions[bot] edited this page Nov 21, 2026 · 3 revisions

The code can be cloned again; the months of editorial judgement can't. That lives in the config, the hand-edit backups, the enrolled faces and voices, the cast photos, the layouts and each stream's review state. clippa backup zips those. clippa streams shows what the stream folders take on the share and frees the big files of streams that are done with.

clippa backup

clippa backup --dry-run        # what would go in, by folder, with sizes
clippa backup                  # write backups/taste/taste-<date>-<time>.zip
clippa backup --with-secrets   # secrets/ too
clippa backup --to D:/offsite  # somewhere else

What goes in, relative to the config file's folder (/data on the NAS hub):

What Where
The config clipper.toml, clipper.local.toml
Hand-edit copies backups/ (not the backup folder itself)
Schedule, people, spend schedule.json, people.json, logs/costs.jsonl
Faces and voices faces/ (not the downloaded models), voices/
Shorts layouts layouts/
Cast photos [thumbnail] photos, with each person's cast.json
Your own prompts [paths] prompts, if set
Each stream config.toml, meta.json, review.json, review_state.json, thumbnails.json, retakes.json
Secrets secrets/, only with --with-secrets or [backup] with_secrets

These stay out: the recordings and cuts (the videos are on YouTube), the transcripts (Modal can redo them), deploy/*/env.local on the machine you deploy from, and Kitsu's database. All of this install's zips together come to about 30 MB.

serve makes one when the newest is [backup] every_days old (7; 0 = only by hand) and keeps the newest [backup] keep (8). clippa doctor and the weekly Audit say when the newest is older than [backup] max_age_days.

A zip on the NAS doesn't survive the NAS. Copy backups/taste/ to another disk now and then. To restore, unzip it over the config folder (or pick out the files you need).

Kitsu

Every review decision and note is in Kitsu's database, in its Docker volumes on the NAS. Stop Kitsu, then tar each volume (docker volume ls lists them):

docker run --rm -v <volume>:/v -v "$PWD":/b alpine tar czf /b/kitsu-<volume>.tgz -C /v .

Disk: clippa streams

clippa streams list                  # size per stream: source, audio, cuts, previews
clippa streams prune                 # what prune would remove (older than 30 days)
clippa streams prune --yes           # remove it
clippa streams prune --cuts --yes    # and the long cuts of uploaded or rejected items
clippa streams prune --stream kick-1 # one stream; says why when it's left alone

A Kino night takes 60-90 GB: the source recording is about 30 GB, the audio 1 GB, the segment and compilation cuts 20-50 GB and the review previews 3-8 GB.

prune removes the source (source.*) and the audio (audio.*). It keeps every JSON file (transcript, picks, review state, config.toml), so the cheap stages and archive lookups still work. --cuts also removes the segment and compilation video and the review preview of each item that is uploaded or rejected. An approved item the scheduler let go keeps its file, so a pin can still send it. Clips and Shorts always stay, because X and TikTok post from them. What was removed goes in the stream's pruned.json, and a stage that needs the source afterwards says it was pruned.

A stream is left alone when:

  • it is in [streams] never_prune (a test stream you re-cut)
  • it is recording, or the worker is running it
  • it isn't in review yet
  • it is younger than --older-than (default [streams] keep_source_days, else 30 days)
  • an item is still waiting on a decision, is scheduled and not uploaded, or is held while it could still go out ([schedule] stale_days after the stream)

With [streams] keep_source_days set, serve runs clippa streams prune --yes once a day. It is 0 (only by hand) by default.

To keep recordings as long as the disk has room, also set [streams] keep_free_pct (say 15): serve's daily prune then takes streams older than keep_source_days only while the share has less than that percent free, oldest first, and stops as soon as it has enough (clippa streams prune --while-free-below 15 does the same by hand). keep_source_days stays the floor: nothing younger goes. The checkup warns when free space is under keep_free_pct (Checkup).

Clone this wiki locally