Repository navigation
v0.6.0
Highlights
AI recommendations (For You). A per-account "For You" feed that ranks your
unread entries with an AI provider of your choice. Save one or more provider
profiles and switch between them, run a ranking on demand or on a per-user
schedule, and follow it with a live run view, an ETA, and cost bounds. A
background worker and a detached drainer carry long runs to completion, and
results are scored so the strongest articles surface first.
Full-text search. Search across your entries, backed by Meilisearch with a
permanent database fallback so search keeps working even when the index is
unavailable. You choose whether to include Meilisearch at install time — the
database fallback covers the rest. Whole-word matching is supported, and results
reload in place without blanking the list.
Account backup and restore. Export a whole account — feeds, tags, read
state, and your kept and favorited articles — and restore it into another
instance. A self-hosted install accepts a real account backup, so you can move
between instances or keep an off-site copy.
Simplified install and update. The production installer sets up the full
stack from a single package question and survives a second run, and
scripts/update.sh moves an existing install to the latest release. Both
resolve the latest release straight from git, with no dependency on the GitHub
Release object.
What's Changed
- Per-account AI provider configuration (groundwork) by @larspohlmann in #306
- Record actively opened entries as viewed (#307) by @larspohlmann in #313
- feat(#308): AI-powered For you recommendation feed by @larspohlmann in #314
- feat(#311): background worker container for recommendation runs, feed refresh and housekeeping by @larspohlmann in #315
- Score-based recommendation ranking (#316) by @larspohlmann in #317
- Stream provider responses in the AI client for early stall detection by @larspohlmann in #318
- Realtime debug view for recommendation runs (#309) by @larspohlmann in #319
- Stop charging reasoning tokens to the answer's budget (#320) by @larspohlmann in #322
- For-you polish: layout, controls, cost bounds (#321) by @larspohlmann in #324
- Move the For You run trigger into the list title bar by @larspohlmann in #326
- fix(#327): reasoning models fail ranking — max_tokens starves the answer by @larspohlmann in #328
- fix(#329): LM Studio json_schema, real transport errors, batch degrade, resume-or-fresh choice by @larspohlmann in #330
- For You: show the recommendation strip in every reading layout (#331) by @larspohlmann in #332
- feat(#333): scheduled auto-generation of For You (per-user cadence + external-cron endpoint) by @larspohlmann in #335
- Save multiple AI provider configurations and switch the active one (#334) by @larspohlmann in #337
- fix(#323): recover the recommendation answer from reasoning_content when content is empty by @larspohlmann in #339
- For-You run: ETA + anticipatory progress bar (#336) by @larspohlmann in #340
- feat(#323): per-config toggle to ask the provider not to reason by @larspohlmann in #341
- fix(#342): mobile layout regressions + settings polish by @larspohlmann in #343
- Parallel recommendation batch calls + recommendation-quality tuning (#344) by @larspohlmann in #345
- feat(#346): single /maintenance/tick endpoint (refresh + recommendations) by @larspohlmann in #349
- feat: duplicate an AI provider configuration + collapsible AI settings UI (#347) by @larspohlmann in #350
- For You: run-boundary headers between recommendation runs (#348) by @larspohlmann in #351
- refactor: collapsible settings-card composes app-disclosure (#352) by @larspohlmann in #353
- fix(#354): pin every resolved IP so the fetch client can fall back across families by @larspohlmann in #355
- fix(#356): cross-family fetch failover on a dead-route reset by @larspohlmann in #357
- fix(#358): cross-family failover on a route-specific error status (taz.de) by @larspohlmann in #359
- fix(#360): force a fresh connection on a failover retry (taz.de) by @larspohlmann in #361
- fix(#362): exclude the worker from the e2e stack boot by @larspohlmann in #363
- fix(#364): tighten adaptive refresh bounds (floor 5 min, ceiling 6 h) by @larspohlmann in #365
- feat(#366): sort entry list by refresh run, not article publication time by @larspohlmann in #367
- fix(#368): read Atom dc:date so tagesschau & NDR entries get a published date by @larspohlmann in #369
- fix(e2e): let the seeded admin accept a scraped candidate by @larspohlmann in #377
- refactor: give the page URL a home in the scraper and discovery by @larspohlmann in #375
- build: gate the backend on phptramp by @larspohlmann in #379
- build: run the gate against phptramp's latest develop by @larspohlmann in #381
- fix(security): php_codesniffer 3.13.6 closes CVE-2026-67434 by @larspohlmann in #383
- fix(#388): follow phptramp to Packagist by @larspohlmann in #389
- fix(#384): sort on a clamped effective date, purge by fetch date by @larspohlmann in #387
- feat(#386): recommendation look-back window (1-7 days, default 2) by @larspohlmann in #391
- Spawn a detached CLI drainer to drive recommendation runs to completion (#371) by @larspohlmann in #392
- feat(#372): in-page documentation for the AI settings by @larspohlmann in #395
- fix(#396): reject a dedup reply that names most of the list by @larspohlmann in #397
- fix(#399): stop the batch prompt asking for candidates to be left out by @larspohlmann in #400
- feat(#401): keep the debug log of the last ten runs by @larspohlmann in #402
- feat(#403): score on 0-1000, and ask for exact values by @larspohlmann in #404
- feat(#406): dedup lines carry the description, not the reason by @larspohlmann in #407
- Search entries by title and summary by @larspohlmann in #413
- Show the entry actions on every magazine card by @larspohlmann in #418
- fix(#415): report a rejected AI configuration where it happened, with the server's reason by @larspohlmann in #420
- feat(#398): show For-You progress in the app-wide pill, not only the reader header by @larspohlmann in #421
- fix(#423): parse feeds that start after the XML declaration by @larspohlmann in #425
- fix(#424): report a bot gate as a refusal, not as a missing feed by @larspohlmann in #426
- docs(#428): target the README at end users, polish the operator guides by @larspohlmann in #429
- feat(#430): print the setup information last, and default to port 3333 by @larspohlmann in #431
- feat(#409): record what each recommendation run costs, and show a month-paged run history by @larspohlmann in #427
- Per-connection AI timeout profile, and the info tip that opened in the wrong place by @larspohlmann in #434
- Design polish sweep by @larspohlmann in #436
- fix(#437): bound a runaway completion where the runaway is by @larspohlmann in #438
- fix(#441): clean up the #437 runaway bounds, and close what review found by @larspohlmann in #442
- Meilisearch-backed entry search with a permanent database fallback (#432) by @larspohlmann in #446
- fix(#448): let a search reload the list without dimming or veiling it by @larspohlmann in #449
- fix(#450): carry the whole-word mode into the search index query by @larspohlmann in #451
- Bound a stranded recommendation run to minutes, split slow_model, move the drainer spawn by @larspohlmann in #452
- docs(#432): move Meilisearch wire-format evidence out of scratch by @larspohlmann in #447
- Back up and restore a whole account between instances (#412) by @larspohlmann in #457
- fix(#458): let a self-hosted install accept a real account backup by @larspohlmann in #459
- fix(#462): fade a list's rows as soon as they render by @larspohlmann in #463
- feat(#453): one package question, with the memory each one needs by @larspohlmann in #464
- fix(#465): fit the run-history row on a phone by @larspohlmann in #466
- fix(#467): keep the image when a page lazy-loads it by @larspohlmann in #469
- Reader header: kept/favorite on mobile, per-article refresh, shared loading overlay (#470) by @larspohlmann in #471
- Box publisher inserts as cards, strip orphaned icon glyphs (#472) by @larspohlmann in #474
- fix(#468): halve the article reading-focus change from #435 by @larspohlmann in #477
- feat(#478): add "Recently read" saved view by @larspohlmann in #479
- fix(#476): reader extracts the wrong block — extract both ways, keep the richer by @larspohlmann in #481
- feat(#482): tick=viewed, circle=read; enforce viewed⊆read by @larspohlmann in #485
- Visual refinements: AI settings hint, mobile search, sidebar chevrons, entry rows (#486) by @larspohlmann in #487
Full Changelog: v0.5.6...v0.6.0