* fix(opds): show 'Open & Read' for publications already in the library
PublicationView only flipped the acquisition button to 'Open & Read'
when the user downloaded the book within the current component
lifecycle — the downloadedBook state started as null on every mount
and was never reconciled against the actual library. So reopening the
detail page (after navigating away in the OPDS browser, or coming back
from the reader) always showed 'Download' / 'Open Access' again and
asked the user to re-download a book they already had.
Two real-world feeds (m.gutenberg.org and ManyBooks) exposed two
distinct failure modes that made naive title/author equality unusable:
(1) Gutenberg's OPDS entries never carry <dc:identifier>. The only work
identifier lives in <atom:id>urn:gutenberg:1342:2</atom:id>, which
foliate-js's opds.js parses into metadata.id — a field
getMetadataHashInfo never reads. So the identifier candidate list
was empty and identifier-overlap matching silently skipped every
book.
(2) Author strings disagree across the boundary: the feed emits
'<author><name>Austen, Jane</name>' (Lastname-first) while the EPUB
inside the same feed ships <dc:creator>Jane Austen</dc:creator>.
Plain string equality (even after normalization) never matched.
Add findExistingBookForPublication in app/opds/utils/findExistingBook.ts
with a layered matcher:
- Pass 1: full metaHash equality (the strongest signal — same fingerprint
the bookService uses internally).
- Pass 2: identifier overlap. collectPublicationIdentifiers() splices
metadata.id into the candidate list alongside metadata.identifier, so
Gutenberg's 'urn:gutenberg:1342:2' feeds into identifierKeys(), which
extracts the '1342' digit-tail (>=3 digits, so the trailing ':2'
version suffix is ignored) and matches the EPUB's
'http://www.gutenberg.org/1342' → '1342'.
- Pass 3: tolerant title + author match. hasAuthorOverlap() does a
token-set fallback: each name is split on whitespace/comma/semicolon,
single-letter tokens (initials) and year-range tokens ('1775-1817')
are dropped, and we require >=2 shared tokens by default. So
'Austen, Jane' ↔ 'Jane Austen' match via {austen, jane} but
'Author A' ↔ 'Author B' don't (both collapse to {author} after
dropping single-letter tokens — we track raw token count to refuse
the single-token shortcut when discriminative parts were filtered).
Genuine mononyms still match on a single shared token because raw
count is 1 on both sides.
OPDSPerson[] is fed through the library's own formatAuthors so the
produced author string matches Book.author byte-for-byte (Intl.ListFormat
output, locale-aware). Soft-deleted books are skipped so removing a copy
locally reverts the button.
Wire it in opds/page.tsx by subscribing to useLibraryStore.library
(rather than the snapshot reads inside handleDownload) and memoizing
existingBookForPublication. Pass it to PublicationView, which now
seeds downloadedBook from the prop and resyncs when the prop changes
(switching publications inside the browser) — but does not clobber a
download success it observed locally before the parent recomputed.
Covered by unit tests for the helper, including the real Gutenberg
URN-in-atom-id case, ':version' suffix on the URN, 'Lastname, Firstname'
vs 'Firstname Lastname', year-range stripping, and the 'Author A' vs
'Author B' non-match.
* fix(opds): dedupe downloads by source url
---------
Co-authored-by: Huang Xin <chrox.huang@gmail.com>
* feat(reedy): wire RetrievalBackend interface + metrics into the chat adapter
Phase 1B backend integration. Adds a RetrievalBackend interface so
TauriChatAdapter holds a uniform reference instead of branching across the
file, with two impls — LegacyIdbBackend (wraps the existing IDB ragService
unchanged) and ReedyBackend (lazy-opens reedy.db, adapts the active
provider's embedding model to Reedy's narrower shape, exposes a Vercel
`lookupPassage` tool). selectBackend() gates Reedy behind both
aiSettings.reedy.enabled AND isTauriAppPlatform() per plan D15 so the MVP
cohort is desktop-only.
The Reedy path streams via streamText({ tools: { lookupPassage }, stopWhen:
stepCountIs(3) }) with a status-aware system prompt that tells the model
how to phrase responses for each RetrieverStatus value.
Replaces the module-global `lastSources` + 500ms poll with a per-instance
ReedySourceStore keyed by a synthetic per-turn id the adapter generates,
so the Sources dropdown stops racing on global state. Both legacy and
Reedy backends now feed citations through the same store; the UI is
backend-agnostic via a shared SourceItem shape both ScoredChunk and
RetrievedChunk satisfy.
Adds the reedy_metrics table to the reedy migration (versioned with
app_version + session_id + turn_id per row) plus a ReedyMetrics writer
that ReedyBackend uses to record indexing-lifecycle and tool-use events.
Always-on local; no network egress. NoopReedyMetrics keeps construction
cheap before the DB opens.
Tests: retrievalBackend selectBackend gates, ReedySourceStore semantics
(append/replace/subscribe/clear), ReedyMetrics debounced batching +
exportBundle, and a TauriChatAdapter contract test that asserts the
Reedy/legacy code-paths pass the right args to streamText.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(reedy): UI wiring — settings toggle, clickable sources, feedback bundle
Phase 1B UI integration. AIAssistant now constructs the active backend
(legacy or Reedy) via selectBackend with the platform gate from M1.7,
wires a ReedySourceStore for the chat adapter, and routes Sources-dropdown
clicks to `getView(bookKey)?.goTo(source.cfi)` when the source has a CFI.
Legacy-path sources still render as static rows because they have no CFI.
AIPanel grows a 'Reedy Retrieval (Beta)' BoxedList with the toggle and a
'Send Reedy feedback' button that calls exportReedyMetricsBundle and
triggers a JSON download of the last 90 days of events. The toggle is
disabled on web with an explanatory description per plan D15.
Thread accepts an onSourceClick callback and renders each source as a
button when its source carries a CFI, otherwise as a static div — so the
Sources dropdown is backend-agnostic via the shared SourceItem shape.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(db): add reedy schema migration with Tantivy FTS + lazy embeddings
Registers a new `reedy` migration set bound to reedy.db. Creates
reedy_book_meta + reedy_book_chunks with a Tantivy FTS index on
chunks.text (ngram tokenizer) and a per-book position index used by
BookRetriever. The vector embeddings table is intentionally NOT created
here — the indexer creates it lazily on first index so the vector32(<dim>)
column matches the active embedding model. Tests cover the migration
applies cleanly, is idempotent, and that the FTS index is queryable.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(reedy): retrieval primitives — DB, chunker, indexer, retriever, lookupPassage tool
Wires the MVP retrieval pipeline behind reedy.db, all under src/services/reedy/
and with no integration into the existing AI module yet (Phase 1B will do that).
- ReedyDb wrapper over DatabaseService: book-meta CRUD, lazy embeddings
table at the active model's dim, bulk chunk + embedding writes via batch(),
hybridSearch (brute-force cosine + Tantivy FTS + reciprocal-rank fusion
with 3× per-path over-fetch), per-book and global wipe. Internal write
queue serializes batch() calls so Turso's single-writer transaction guard
doesn't trip when BookIndexer runs across books in parallel.
- CfiChunker: TreeWalker over the section's DOM, ~maxChunkSize windows with
paragraph > sentence > word break-points, full epubcfi(/6/N!/…) anchors,
round-trip verified via CFI.toRange before each chunk lands.
- BookIndexer: per-book mutex, lazy embeddings-table creation, model.batchSize
embedding batches with dim assertion, terminal status transitions
(indexed | empty_index | failed). Re-indexing clears prior chunks via a new
ReedyDb.clearBookChunks helper.
- BookRetriever: status-typed results (ok | not_indexed | empty_index |
stale_index | degraded). Embedding has a 5s wall-clock budget; on timeout
it falls through to FTS-only with status=degraded.
- lookupPassage Vercel ai-SDK tool: Zod-validated query/topK, per-turn
composite-key dedupe, parallel-call serialization, 10s per-turn budget,
6000-char result clamp, status-with-hint passthrough, and a separate
serializeForModel that wraps each passage in <retrieved trust="untrusted">
with XML-escaped content so book text cannot escape the envelope.
59 unit tests; pnpm test + pnpm lint clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(db): forward DatabaseOpts to tauri-plugin-turso
NativeDatabaseService.open ignored its opts parameter, dropping any
experimental feature flags (e.g. 'index_method' needed for FTS / vector
indexes) and any encryption config before they could reach the Tauri
plugin. Translate DatabaseOpts to the plugin's LoadOptions shape and
forward as the single argument Database.load accepts. Skip translation
when no relevant opts are set so the existing path-string call shape is
preserved for callers without experimental/encryption needs.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(db): cover Turso vector primitives + add benchmark harness
Verifies the Turso functions Reedy retrieval depends on, with a
brute-force per-book kNN test that runs against every DatabaseService
backend (node, native, WASM):
SELECT vector_distance_cos(embedding, vector32(?)) AS d
FROM book_chunks
WHERE book_hash = ?
ORDER BY d ASC LIMIT k
This is the path Turso's own founder recommended in
tursodatabase/turso#3778 ("First, focus on efficient SIMD-accelerated
brute-force search") and what shipped at commit 1aba105df4f. Native
vector index modules don't exist in this engine: `libsql_vector_idx`,
`vector_top_k`, and `USING vector/hnsw/diskann/ivfflat` all parse-error
against @tursodatabase/database@0.6.0-pre.28 (libsql_vector_idx is a
libSQL/sqld fork feature; DiskANN was closed not-planned upstream in
#832). The test asserts cross-book isolation and nearest-first ordering
using only `WHERE book_hash = ?` and `ORDER BY` — no DDL, no identifier
interpolation, no index plumbing.
Also adds bench/ harness for manual perf checks:
pnpm bench [name] run benchmarks (refuses in CI)
pnpm bench --list list available benchmarks
pnpm bench --no-record skip results.jsonl append
pnpm bench --force override the CI guard
Uses Node 24's --experimental-strip-types so no tsx devDep is needed.
Appends one JSON line per run to bench/results.jsonl (gitignored, local
history; share by pasting tabular stdout into PRs/issues). Explicitly
NOT in CI — shared-tenant variance makes synthetic-benchmark regression
detection unreliable; production telemetry (reedy_metrics, plan §M1.9)
is the right tool for that.
First benchmark: vector-retrieval. Measured on M1 Pro:
400 chunks × 384 dim → 0.35 ms / query
400 chunks × 768 dim → 0.45 ms / query
2000 chunks × 768 dim → 2.23 ms / query
10000 chunks × 768 dim → 14.00 ms / query
400 chunks × 1536 dim → 0.70 ms / query
Per-chunk cost ~1.1 µs at 768 dim = ~1.4 ns/dim. NEON-class on
Apple Silicon, ~50× faster than scalar — confirms SIMD acceleration is
active in 0.6.0-pre.28. Per-query latency stays sub-ms at Reedy MVP
corpus sizes; the ceiling is ~10K chunks per book before phone-class
hardware notices.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(hardcover): add one-way Hardcover sync integration
- Add HardcoverClient with ISBN/title-based book lookup, progress push, and note sync
- Add HardcoverSyncMapStore for persistent local mapping of note IDs to Hardcover journal IDs
- Add GraphQL queries/mutations for Hardcover API (insert/update/recreate journal entries)
- Add DB migration for hardcover_note_mappings table (schema: hardcover-sync)
- Add HardcoverSettings dialog component (connect/disconnect/enable toggle)
- Add useHardcoverSync hook wired in Annotator for push-notes and push-progress events
- Add Hardcover Sync section in BookMenu with per-book toggle (persisted in BookConfig)
- Add HardcoverSettings type and DEFAULT_HARDCOVER_SETTINGS to system settings
- Add hardcoverSyncEnabled per-book flag to BookConfig
- Mount HardcoverSettingsWindow in Reader alongside KOSync and Readwise windows
Sync is one-way (Readest → Hardcover), manual-action only, per-book opt-in.
Supports idempotent note sync: insert new, skip unchanged, update changed, recreate on stale remote ID.
* fix(hardcover): proxy API calls server-side to fix CORS, normalize Bearer token
- Add /api/hardcover/graphql Next.js route that proxies POST requests server-side,
bypassing CORS restrictions (Hardcover API has no Access-Control-Allow-Origin header)
- On Tauri (desktop), calls Hardcover directly; on web, routes through the proxy
- Normalize token in HardcoverClient: accept raw JWT or 'Bearer <jwt>' format
- Update helper text to point to hardcover.app → Settings → API
* fix(hardcover): surface note-sync no-ops and harden book resolution
- Show toast feedback when book data is still loading, Hardcover is not configured,
or the current book has no annotations/excerpts to sync
- Show an explicit info toast when note sync finds no new changes
- Parse Hardcover search results in the current hits/document response shape
- Resolve note sync through ensureBookInLibrary for parity with progress sync
- Add console logging for note/progress sync failures
* debug(hardcover): add runtime instrumentation for note sync
* feat(metadata): keep identifier stable and store ISBN separately
- Add dedicated metadata.isbn field
- Expose ISBN as its own editable field in book metadata
- Preserve identifier semantics for existing source IDs and hashes
- Route metadata auto-retrieval ISBN handling through the new field
- Prefer metadata.isbn for Hardcover matching
* fix(hardcover): avoid wasm sqlite for note mappings on web
- Store Hardcover note sync mappings in localStorage on web
- Keep sqlite-backed mappings for desktop/native environments
- Remove the web-only database dependency from manual note sync
* fix(hardcover): dedupe notes by payload hash across unstable note IDs
- Add payload-hash lookup in HardcoverSyncMapStore
- Reuse existing journal mapping when payload already synced
- Prevent duplicate insertions when note IDs change or duplicate locally
* fix(hardcover): avoid duplicate quote export when annotation note exists
- Detect excerpt+annotation pairs for the same highlight
- Skip standalone excerpt export when annotation has note text
- Keep annotation export as the single source of truth
* docs(hardcover): add consolidated change summary for review
* fix(hardcover): suppress excerpt export by CFI when annotation note exists
* fix(hardcover): suppress empty-note annotation duplicates when note exists
* fix(hardcover): deduplicate notes by text and cfi base node
* refactor(hardcover): optimize sync performance, add rate limiting and clean up debug tools
* chore: remove dev-only change log from main
* test(hardcover): add unit tests for sync mapping and client logic
* chore: custom deployment and UI fixes
* fix(hardcover): use timestamptz for accurate annotation time
* fix(hardcover): use date scalar and RFC 3339 formatting for journal entries
* Revert "chore: custom deployment and UI fixes"
This reverts commit 0329aba7129d1e1ebf2c663804b8fba9a9f87b91.
* Fix hardcover progress dates and surface imported ISBNs
* Fix Hardcover currently reading sync
* fix(hardcover): avoid promoting note sync status
* style(hardcover): apply prettier formatting
* test(hardcover): fix strict TypeScript assertions
* test(hardcover): harden sync regression coverage
* test(hardcover): fix lint and formatting regressions
* fix(hardcover): narrow note dedupe range matching
* refactor(hardcover): extract note dedupe helpers
* refactor(isbn): extract metadata normalization helpers
* feat(hardcover): improve synced quote formatting
loveheaven