CHANGELOG.md

Changelog

All notable changes to Programmarr are documented here. This project follows Semantic Versioning and the spirit of Keep a Changelog.

[0.8.0] — 2026-08-19

Hardening release ahead of going public. Programmarr had only ever run against its author's Plex and Tunarr; this release is the result of auditing it for everything that breaks, misleads, or silently does nothing on somebody else's setup.

Added

• Tunarr basic auth support. Tunarr gained optional HTTP basic auth on its API; Programmarr sent no credentials at all, so a locked-down Tunarr failed everywhere. New tunarr_username / tunarr_password config keys, editable in Settings → Connections (and in onboarding). Every Tunarr call in every module now goes through one shared header builder, so enabling auth can't half-work.
• Test connections in onboarding (POST /api/test-connection). Saving config never contacted Tunarr or Plex, so a typo'd URL produced a green "Setup complete" and a broken install. The wizard now validates what you typed before writing it. A failed test warns but still lets you continue — it informs, it doesn't gatekeep.
• Pre-delete lineup backups. Deploy can delete Tunarr channels, which is irreversible. Programmarr now writes a gzipped snapshot of everything it's about to remove — each channel plus its full programming — to tunarr_backup_*.json.gz in your data directory before the first delete. The 3 most recent are kept; dry runs write nothing. See the new Backups section in the README.
• Tunarr version reporting. The Dashboard's connection status now includes Tunarr's self-reported version, so bug reports can state it without digging.
• Issue templates and Discussions, so bug reports arrive with the version and log information needed to actually act on them.

Fixed

• Channel numbering order was ignored by the AI path. PROMPT.md had been rewritten to a numbered list while the code still matched the old wording, so the function that rewrites the numbering scheme was silently doing nothing — the LLM always received the default category order regardless of Settings → Channel Numbering. A non-match now warns loudly instead of passing the content through unchanged.
• Pre-1970 films were invisible to the decade facet. Decade buckets started at 1970, so a classic-film collection contributed nothing at all. Buckets now run from the 1920s. (On the author's library this made 132 previously-unreachable films eligible for decade channels.) New labels are spelled in full — "1950s", not "50s", which would be ambiguous beside "2020s". Existing 70s/80s/90s labels are unchanged, so current channel names and saved planner state keep matching.
• A failed library fetch looked like an empty library. Every API error collapsed to "no results", so an outage reported "Indexed 0 movies" and would deploy channels with nothing on them. Failures are now distinguished from genuinely empty libraries: all libraries failing raises, a partial failure warns and continues.
• Misleading errors for unsupported setups. "No Plex source found in Tunarr" covered four unrelated situations and sent Jellyfin users off debugging Plex. Now each is named specifically: an unreachable Tunarr, a Tunarr with no media sources, a non-Plex source ("Found: jellyfin"), a Tunarr too old to have the required endpoint, and — separately — credentials that are missing versus credentials that are wrong.
• A broken pipeline run hung the UI forever. The progress stream ends on a completion event; anything failing after the response started (a subprocess that won't launch, a dead pipe, an unwritable log) killed the stream with no error and no exit code, leaving a spinner. Every failure path now terminates properly.
• An empty library picker with no explanation. A Plex failure while listing libraries was swallowed by a comment claiming the UI showed the error elsewhere. It didn't.
• A wrong Plex token could pass the connection test. Many self-hosted Plex servers allow unauthenticated access on the local network, so every local request succeeds — even with no token. That result is now flagged as unverified rather than reported as a validated token.
• Empty Planner sections now say why. "No TV channels available — run Export" was shown for three different situations, including to users who had already exported. A movies-only library is now told plainly there are no TV channels to build; a library with shows that don't qualify gets the actual thresholds named.

Changed

• Deploy-time backups are gzipped and retention is 3 rather than 10. Measured on a real 156-channel server: 151 MB raw, 12 MB compressed — at the old retention that was 1.5 GB quietly accumulating in a data directory.
• README quick start now includes the compose file inline; previously only the TrueNAS section had one, so anyone who hadn't cloned the repo had nothing to run. Requirements now state Tunarr 1.x + Plex explicitly, and that Jellyfin/Emby are not supported.

Notes

Plex remains required. If you'd like Jellyfin or Emby support, add a reaction to #36 — demand decides priority.

[0.7.1] — 2026-06-20

Fixed

• Title-collision resolution. When a library holds a movie and a TV series with the exact same title (e.g. the 2017 Baywatch film and the 1989 series), resolve_title picked the movie first and returned unconditionally — so a marathon channel could loop a single film instead of the series. Resolution now prefers whichever copy has more playable programs (the same tie-break build_library_index uses for duplicate shows): a real series beats a lone movie; an all-missing series still yields to the movie. Applies to every path that resolves titles (initial deploy, live re-resolve, surgical/apply).

Changed

• Dependency cleanup. Dropped unused @mantine/code-highlight (frontend) and aiofiles (backend) — neither was imported anywhere.

[0.7.0] — 2026-06-11

Added

• Playback structure (per-channel). Channels can carry "playback": {"structure": "interleaved"|"timeline", "episodes_per_block": N}. Interleaved keeps movies in watch order with ~N-episode blocks between them; timeline posts a manual Tunarr lineup in strict release order (show runs air at their premiere position). The field is plumbed through every deploy path (compose, surgical deploy, nuke, apply) and editable per channel in the Channels editor. Live franchise channels default to interleaved/4.
• Franchise live content-refs. A new identity-based ref {"match": "franchise", "name": "…", "order": "release_date", "exclude": []} lets live channels auto-add franchise members that don't share a name (resolved from the cached TMDB belongs_to_collection + Wikidata franchise index, never name matching). Authored by the Planner's per-franchise "Keep updated" switch; the scheduler, surgical deploy, and apply all resolve it; recipes preview supports it.
• Channel icon overhaul. Every channel now gets an icon: verified TMDB logos where trustworthy (kind-gated, name-must-match-query) and generated name-stamped badges everywhere else (badge_renderer + committed badge_assets/). New Channels-editor icon control (badge / TMDB / custom upload / reset-to-automatic) via POST /api/channels/{number}/icon, with pin support so manual icons survive the automatic art pass. tmdb_api_key is now optional — without it everything badges.
• Multi-source / multi-server library support. Export auto-supplements from all Tunarr media sources; the Export-step library picker shows every source grouped by server. Multi-Plex-server support across export, deploy, and Settings.
• Planner refinements. Deselect-all on entity dropdowns, a reset confirmation, Nuke now clears candidate selections, and auto search+scroll for large candidate lists.

Fixed

• Icons never use SVG logos. Plex guide icons must be raster — SVG TMDB logos are now rejected so the guide renders.
• Export-step library labels. Divider labels group library checkboxes by source server; redundant per-label server suffix removed.

Changed

• Self-healing dev loop. dev.ps1 kills stale instances before starting.

0.6.0 — 2026-06-10

Added

• Multi-title channel logos. fetch_images.py now fetches real TMDB logos for network, franchise, entity (director/actor), and generic multi-title channels — not just solo-title channels. Kind-aware search strategies: network channels try company/network search first; franchise channels try the cleaned name then the first content item; director/actor channels use the cleaned name; everything else falls back through TV → movie → company. Kind hints are read from planner_state.json; missing file is a safe no-op. Solo-title path is unchanged.
• In-app update banner. When a new GitHub release is available, a dismissible banner appears in the app linking to the release notes. Backed by GET /api/update-check, which polls the GitHub Releases API and caches the result.
• "Check for updates" toggle in Settings. Update polling is on by default; users can opt out without rebuilding the container. Persisted to config.json as update_check_enabled.

Fixed

• Live-channel name-match guard. update_channel_in_place now takes an expected_name and refuses to patch (raises, then skips + logs) if the Tunarr channel at that number carries a different name. Prevents silent content scrambling when two Programmarr instances share one Tunarr server — or when channels.json drifts out of sync with Tunarr's numbering. Wired into the scheduler, the Channels-page "Save and Apply," and the surgical deploy.
• Live-channel indexing now covers all Plex libraries. build_library_index previously indexed only the first enabled movie and first enabled TV library, silently dropping entire secondary libraries (e.g. a "Cartoons" section alongside "TV Shows"). It now aggregates every enabled library of each kind.
• Live-channel dedup across libraries. A show that appears in more than one Plex library is indexed once, preferring the copy with the most playable (non-missing) episodes. Without this, a dead duplicate could shadow the real copy, or doubled episode counts caused a permanent diff mismatch that churned the channel every cycle.
• Light mode theming. Introduced semantic CSS surface/border tokens (--app-bg, --surface-panel, --surface-sunken, --border-subtle) defined per color scheme, replacing 40+ hardcoded dark values across 7 components that never flipped in light mode. Warm off-white page background replaces blinding gray-0. Terminal surfaces (SSE output, Logs, AI prompt) stay dark in both modes by design. Dark mode is visually unchanged.
• Logs page readable in light mode. The log code block forces a dark terminal background but text was inheriting Mantine's default dark color — dark-on-dark, unreadable. Text is now pinned to the same light terminal value used by the SSE terminal and AI prompt preview.

[0.5.0] — 2026-06-09

A large Planner overhaul: smarter channel discovery from new data sources, a guided three-section Planner, sticky picks, and a safe surgical add/edit deploy.

Added

• Three-section Planner (TV / Movies / TV+Movies). A single-open accordion replaces the long candidate list. Categories open by default with a per-category Done collapse; a new TV+Movies section builds mixed-genre channels that interleave episodes and films of a shared genre.
• Sticky Planner picks. Your selections persist to planner_state.json (saved on every change, restored every run) — reopen the Planner and your last lineup is already checked. Nuke keeps your picks (it only wipes Tunarr); Clear all is the sole reset.
• Add / Edit vs Nuke. A top-of-Run choice. Add/Edit does a surgical deploy — creates new channels, deletes unchecked ones, updates changed ones in place (preserving the Tunarr id + Plex DVR mapping), and never touches live or hand-authored channels. Nuke wipes and rebuilds from channel 1.
• Deploy diff preview. Before anything touches Tunarr, see exactly what will change —
  • new / ~ changed / − removed / = unchanged / · not-managed — then Confirm.
• Real networks from TVmaze (HBO, NBC, Apple TV+…) — replaces the unreliable Plex "studio" field, no API key needed.
• Franchise detection spanning TV + film, from TMDB + Wikidata (per-member checkboxes — pick your favorite entries). Runs as a cached background scan with a progress bar.
• Themed channels from a curated TMDB-keyword catalog (Heist, Time Travel, Christmas, Dystopian, Superhero, …) — deterministic, no AI step needed.
• Classic TV programming blocks (TGIF, Must See TV, Saturday Morning Cartoons, …) matched against your library from a built-in catalog.
• Country / Mood / Style channels from Plex metadata (e.g. French Cinema, Film Noir).
• Dashboard: open-in-new-tab buttons for Tunarr and Plex.
• Settings: drag-and-drop reordering of the channel-numbering categories.

Changed

• Channel numbering is now sequential from 1, grouped by category. No fixed block sizes or caps — the lineup is exactly as long as what you pick, and the category order is configurable (Settings + first-run onboarding). Replaces the old channel_blocks sizes with channel_order.
• Collections stay in their own dedicated feature — they're no longer pulled into the Franchise step.

Fixed

• Surgical deploy now records the right channel numbers. Editing an existing channel updates it in place at its real Tunarr number instead of mislabeling it in channels.json.
• Dev loop export works on Windows. The dev backend reloads at the process level (watchfiles) instead of uvicorn --reload, which forced the Selector event loop and broke pipeline subprocesses (export.py etc.). No-op on Docker.
• Dashboard guide no longer repeats the channel number; the auto-update card is trimmed to status
  • last-run.

[0.4.4] — 2026-06-08

Fixed

• Planner now deploys what you built — AI channels and collections no longer vanish. The Planner assembles its lineup in channels.draft.json (compose + the AI "Bring in AI" step + the Collections step all write it), but the Deploy step was reviewing/deploying the wrong file. Two bugs, both fixed:
  • The Deploy review was built from create.py --probe, which read the deployed channels.json instead of the draft — so AI-discovered channels and added collections never showed up and never deployed. The probe now targets the draft (and forwards kept channels as --protect for an accurate delete preview).
  • The Collections step sized its channel numbers from the deployed record, not the draft, so collections landed on top of the AI channels and apply_collections deleted them. A new GET /pipeline/draft lets the Collections step place itself above the full built lineup, preserving every channel.
• Export no longer crashes on a Windows host. Pipeline subprocess output is now forced to UTF-8 (PYTHONIOENCODING), so a title containing a non-cp1252 character (e.g. ⧸) in the unsynced-titles list no longer kills export.py with UnicodeEncodeError. No-op on Docker.

[0.4.3] — 2026-06-08

Fixed

• Dashboard guide now has a single unified vertical scroll. The EPG guide grid previously showed two independent vertical scrollbars — one for the channel rail (number/name) and one for the programme grid — so scrolling one left the other behind. The rail's scroll is now driven by the programme grid via the existing scroll-sync mechanism, so the two move together as one. The mousewheel still works while hovering the channel names.
• AI Extras "Copy" button works over plain HTTP. Copying the AI prompt silently did nothing on non-secure origins (e.g. the production box over HTTP), because navigator.clipboard is undefined outside a secure context and the call threw with no fallback. Added a clipboard helper with a textarea + execCommand('copy') fallback and success/failure notifications, used by both the prompt copy and the XMLTV-URL copy.

[0.4.2] — 2026-06-08

Added

• EPG Guide grid on the Dashboard. The Dashboard now shows a real TV-guide grid (scrollable, 3-hour window, now-line) sourced from Tunarr's XMLTV feed. Clicking any channel in the left rail navigates to its editor on the Channels page.
• Channels page sourced from live Tunarr. The Channels list now pulls from the live Tunarr API so the list always reflects what's actually deployed — no more stale data. Tunarr channels with no channels.json entry show as "Not managed by Programmarr" (read-only orphans).
• Save and Apply (POST /api/channels/{number}/apply). Edit any managed channel and push the change to Tunarr in a single click — updates in place, preserving the Tunarr channel id and Plex DVR mapping. Acquires the same lock as the live-channel scheduler to avoid races.
• Configurable block sizes. Each channel category (Marathons, TV Blocks, Movies, Franchise, Specialty) now has a configurable size in Settings, so large libraries can scale a block without colliding into the next. Channel numbering now defaults to starting at 1 (previously 10). channel_blocks.py is the new single source of truth, shared by the Planner and generate_no_ai.py.

Changed

• Anti-drift deploy model. Planner-flow operations (Compose, Validate, AI Extras, Collections) now write to channels.draft.json instead of channels.json. The deployed record (channels.json) is written only on a successful deploy-selective run, keeping it in sync with what's actually in Tunarr. Abandoning a creation can at worst leave a stale draft, never corrupt the deployed record.

[0.4.0] — 2026-06-07

Added

• Commercials. Any channel can play commercials in the gap between shows — real-TV style. Turn it on with the new 📺 Add commercials toggle in the Planner (applies to every channel you build) or per-channel in the channel editor, and pick which Tunarr filler list to pull clips from. A short pad after each program opens the gap; Tunarr's filler picker fills it (varied selection, no back-to-back repeats). Density self-adjusts — a break between movies on a movie channel, between episodes on a TV channel. channels.json gains an optional per-channel commercials object. (Mid-roll ads inside a show were investigated and deliberately left out — they don't stream on hardware-accelerated Tunarr; see docs/tunarr-commercials-findings.md.)
• Auto-update toggle in the Planner. A 🔄 Keep channels fresh switch marks the channels you build as live, so new episodes and matching films appear on their own as your library grows (runs on the live-channel schedule).
• Advanced config (optional, config.json). tunarr_channel_group and tunarr_stream_mode set the Tunarr group/folder and streaming mode for generated channels. See README → Advanced Configuration.
• Filler-list picker endpoint — GET /api/tunarr/filler-lists.

Fixed

• Settings saves no longer clobber config-file-only keys. config.json is now merge-written, so hand-edited keys (the advanced ones above, and the live-channel recipes_* keys) survive a save from the Settings form instead of being dropped.

Changed

• Export step warns about filler libraries. A clearer, more visible note to leave commercials / trailers / bumper libraries out of the content scan — Plex labels them as "movies," so they'd otherwise sneak into your channel candidates.

[0.3.1] — 2026-06-05

Fixed

• Multi-library exports no longer silently drop content. Two related bugs in export.py where content vanished from plex_library.csv without any error:

  • Tunarr cross-reference checked only the first library of each type. With more than one enabled Tunarr movie (or TV) library — e.g. a main "Movies HD" plus a "Crossovers HD" — only the first was loaded, so everything in the rest failed the cross-ref check and was skipped. It now unions all enabled libraries of each type, and prints per-library counts plus a combined total.
  • CLI auto-detect picked only the first Plex section of each type. Running export.py without --movie-sections/--tv-sections and with multiple Plex libraries (e.g. "TV Shows" + "Cartoons") silently ignored the extras. Auto-detect now collects every matching section. The explicit --*-sections flags and the UI (which always passes keys) were unaffected.

  No CSV schema or downstream changes — a single-library export is byte-identical.

[0.3.0] — 2026-06-05

The Planner — a ground-up rebuild of how you create channels

The biggest release since the web app shipped. The old three-tab Run pipeline (AI / No-AI / Collections) is gone, replaced by a single unified Planner that composes curated, hand-programmed-feeling channels from your library — deterministically — with AI demoted from "the engine" to an optional layer on top. The result feels like a TV lineup a person with taste assembled, not a wall of Comedy Movies (140) blocks.

Added

• The Planner: ingredients → curated candidates. Pick which genres and decades are "in play," then check the exact channels you want from a live, counted candidate list. Everything is built by precise filters — no AI required, no guesswork:
  • Per-show TV Marathons — one 24/7 channel per show; every show in your library is offered, sorted by episode count.
  • TV genre Blocks — themed multi-show channels (Comedy TV, Crime TV…).
  • Genre × Decade — tight era cuts like "90s Comedy" and "80s Horror," nested under each decade.
  • Named Sub-genres — a curated, recognizable set (Rom-Coms, Dark Comedies, Horror Comedies, Crime Thrillers…) instead of arbitrary genre pairs.
  • Studio / Director / Actor channels — A24, "Directed by Tarantino," "Adam Sandler Movies" — the most curated-TV channels there are.
  • Bulk picking — "Top 10" and "Add all" on every section, and sections that fold away once handled so the list never becomes a wall of checkboxes.
• Optional AI layer (✨ Bring in AI). Two things filters can't do, both seeded with your built lineup and merged on top via a copy-prompt / paste-back hand-off:
  • Discovery — the AI suggests themed channels your filters miss (Heist Films, Courtroom Dramas, Time Travel, Sports Underdogs…), and is told what you already have so it never duplicates it.
  • Tonal curation — flag a broad pool (a ✨ on any genre or decade pick) and the AI splits it by tone (Feel-Good vs Raunchy vs Dark Comedies) — the "one comedy doesn't blend into the next" fix that genre tags alone can't make.
• New library metadata for entity channels. export.py now captures Studio and the top-3 billed actors per title (both already present in Plex's response), so studio/director/actor channels work out of the box.
• Facets API. One pass over your library powers the entire candidate list with live counts — genres, decades, the genre×decade matrix, sub-genre blends, and the studio/director/actor/marathon lists.

Changed

• One unified flow: Setup → Export → Planner → [AI Extras] → [Collections] → Deploy. The generation "method" is no longer three separate tabs; it's a single Planner, with Collections-only kept as a quick path.
• Every decision is up front. Keep-vs-wipe of existing channels, whether to add collections, and whether to fetch art are all answered on the first screen — the rest of the flow is review-and-go.
• Deploy is one cascade. It auto-probes on entry, you review/renumber, hit Deploy once, and channel art + Plex sync run automatically — with an always-completes status summary.
• Soft-block channel numbering. Marathons ~10s, TV blocks ~20s, movie channels ~30s+, entity channels ~50s+ — assigned sequentially and spilling gracefully instead of overflowing fixed 20-slot blocks.
• Clearer AI hand-off. The walkthrough now spells out "paste the JSON only," and the prompt is generated from your exact picks.

Fixed

• Merge dedup. Pasting AI suggestions skips any channel whose name already exists, so re-running the AI step can no longer stack duplicate channels.
• Collections numbering. Collections now append above your composed lineup rather than at a fixed ~80 block, so they never overwrite the channels you just built.
• Prominent Plex-sync fallback. When Plex can't auto-add the DVR (common in many setups), the end-of-run summary now shows exactly what to do — with a one-click-copy XMLTV URL — instead of a quiet status line.

Notes

• The deterministic Planner is a strict superset of the old No-AI path. The standalone scripts (generate_no_ai.py, generate_from_collections.py) and the programmarr.py CLI remain for power users.
• Live Channels (v0.2.x) is unchanged and fully compatible — channels built in the Planner can still be marked "live" to auto-update as your library grows.

0.2.3 — 2026-06-04

Follow-up to the Live Channels release: per-channel visibility and control that the original plan called for but v0.2.2 only partly delivered.

Added

• Per-channel "last synced" badge. Each live channel row on the Channels page shows when it was last checked ("synced 2h ago" / "never"). Backed by a new cosmetic data/recipe_state.json — written by the scheduler, read by the UI. It is not correctness state (the change-detection diff still reads live Tunarr); it lives in its own file so it never races with channels.json edits or deploys.
• Per-channel "Save & Sync now". The channel editor (for live channels) gains a button that saves the channel and immediately runs a scheduler cycle scoped to just that channel (/api/recipes/run?only=N), applying the recipe to Tunarr in place without leaving the editor.
• "Next run" on the Dashboard. The Auto-Updates card now shows when the next automatic check will run (e.g. "next ~in 11h"), alongside the last check.

Changed

• /api/recipes/status now returns next_run_seconds and a per-channel channels map. The scheduler's last-run timestamp moved to wall-clock time so the status endpoint can compute the next run without touching the event loop.

Notes

• Documentation reconciled against the full original plan: source/target agnosticism (Jellyfin source, ErsatzTV target) is captured as an explicit future goal with the in-place-update contract noted, but intentionally not built yet — there's no second source/target to validate an adapter layer against.

0.2.2 — 2026-06-04

Added — Live Channels (auto-updating channels)

The headline feature of this release. Channels can now be self-maintaining: mark one "live" and Programmarr re-checks it against your library on a schedule and patches it in place as the library grows — no redeploy, no manual edits.

• Live toggle per channel. In the channel editor, an "Auto-update on a schedule" switch marks a channel live. Live channels show an orange badge in the list. A channel without the flag behaves exactly as before.
• Franchise auto-match. A new content rule, title_contains, pulls in every library title matching a phrase (e.g. Bad Boys → all four films, in release order) and auto-adds new sequels as they appear. Matching is word-boundary, not raw substring, so It matches "It Follows" but never "Little Women". The editor shows a live preview of exactly what matches before you save, with per-title checkboxes to exclude false positives.
• In-place updates — the critical guarantee. Updates reuse the existing Tunarr channel via its id (same channel number, same internal id), so Plex's Live TV / HDHomeRun DVR mapping is never disturbed. Channels are never deleted and recreated for an update. Verified end-to-end: a live Batman channel grew from 162 to 254 programs with its channel id byte-for-byte unchanged.
• Change-detection, not blind re-posting. Each cycle diffs the freshly resolved program set against the channel's current Tunarr programming and patches only on a real difference. An unchanged channel is a cheap no-op — idempotent, with no needless Plex guide churn. There is no state file; Tunarr is the source of truth, so the scheduler survives container restarts for free.
• Background scheduler. A lightweight asyncio loop runs inside the existing FastAPI app (no extra process). It re-reads config every minute, so enabling, pausing, or changing the interval takes effect without a restart. Ships off.
• Controls & visibility.
  • Settings → Live Channels: master enable switch + check interval (hours).
  • Dashboard → Auto-Updates card: on/paused state, live count, last-cycle summary with per-channel changes, a Pause/Resume toggle, and a "Check now" button that runs a cycle on demand.
  • A rolling diff log is written to data/logs/recipes.log.
• New API endpoints (/api/recipes/*): preview (author-time match preview), status, run (apply=false for a safe dry run), pause, and config (merge-writes the enable flag + interval without clobbering connection settings).

Changed

• create.py refactor. The Tunarr resolution logic (library indexing, title/ collection/franchise resolution, schedule building, in-place programming updates) was extracted into a new, dependency-free, importable channel_engine.py. create.py is now a thin CLI wrapper around it, and the scheduler and preview endpoint share the exact same resolver. No behavior change to existing deploys.
• Deploy/scheduler safety. Manual deploys and the scheduler now share a lock so a deploy and an auto-update can never touch Tunarr at the same time.

Fixed

• Channel editor "bounce" on Save. Saving a channel could re-open the modal instead of closing it (and fire the "saved" notification twice) because the list reload re-triggered the deep-link open. The deep-link logic is now guarded so a reload never re-opens a modal you just closed.

Misc

• Added a favicon (orange broadcast icon matching the navbar logo).

0.2.1 — 2025

Added

• Channel protection — choose which existing Tunarr channels to keep before a deploy; protected channels (and their custom logos) are preserved while the rest are rebuilt.
• Channel offset — set a starting channel number so new channels don't collide with ones you keep; auto-calculated from the highest protected channel.

0.2.0 — 2025

Added

• Library picker on the Export step — choose which Plex libraries to scan, and mix multiple libraries of the same type.
• Run pipeline UI rewrite — a premium stepper with export stats, probe review, and selective deploy across the AI / No-AI / Collections paths.
• UI screenshots in the README.

[0.1.x] — 2025

Added

• Initial web app: FastAPI + React + Mantine, served as a single Docker container on port 7979, with a first-run onboarding wizard.
• AI / No-AI / Collections pipeline paths with live streamed terminal output.
• TMDB channel-logo fetching, Plex DVR sync, optional HTTP basic auth.
• Dark / light / auto theme switching.
• GitHub Actions CI publishing to GHCR; Watchtower auto-update support (including the DOCKER_API_VERSION fix for newer Docker engines).