Skip to content

Roadmap

Original goals (plan approved 2026-07-08): (1) iPad interop without leaving reMarkable land, (2) fixed standard page size in landscape, (3) pen strokes that export the way they look on-device, (4) hands-off ~1-2 min sync, (5) notes readable in a normal PDF app, auto-updating.

Phase 0 — XOVI landscape experiment on-device ⏳ WAITING ON DEVICE

Time-critical: the Paper Pro ships for warranty repair (ghosting) around 2026-07-12. Kit is ready in device-mods/ — follow its README checklist: root (dev mode wipes; cloud restores) → remagic installs XOVI → install disableInfiniteScroll + createPagesPaperProSize qmds → THE pivotal test: does clamping the canvas make landscape exports uniform? → iterate qmd if not → rollback.sh BEFORE disabling dev mode → ship. Also while rooted: grab-corpus.sh (test data), pull /usr/share/remarkable/templates/*.svg (exact template art — our dot spacing is approximated), screenshot pen strokes (fidelity references), test whether ghostbuster's full refresh clears the ghosting (repair-claim diagnostic). If a replacement unit returns, re-run install.sh.

Phase 1 — Python core ✅ DONE (2026-07-08)

Library reader, faithful renderer (see docs/formats/remarkable.md), incremental mirror to iCloud Drive (34 notebooks: 68s full, <1s incremental), TOML config, CLI, launchd watch daemon (installed and running). The maintainer confirmed output "looks really correct".

Known gaps / follow-ups: - Pencil / mechanical pencil / paintbrush / shader alphas are best-guess — not exercised by the validation notebook. Calibrate with an ops-diff against an official export of a notebook using them (method in docs/formats/remarkable.md). - Template art approximated (dots ✓, lines/grid rough); replace with real template SVGs from the device (Phase 0 grab). - Annotated PDF/EPUB base-page merge not implemented (scope.pdfs/epubs default false). Plan: pikepdf overlay of rendered annotations onto base. - normalize = "paginate" option (split tall grown pages into multiple fixed pages) not implemented; current default scales to fit. - Typed-text blocks in notebooks render nothing yet (rmscene exposes them; rmc's text layout is a reasonable starting point).

Phase 1.5 — universal converter (M1) ✅ DONE (2026-07-09)

Generalized the reMarkable-only pipeline into a neutral intermediate representation so any ink format can convert to any other. See docs/ir.md for the model, docs/reverse-engineering.md + docs/corpus-protocol.md for how the undocumented formats were (and will keep being) decoded, docs/validated-writes.md for write-safety policy.

  • IR (core/src/inkterop/ir/): neutral document/stroke model, per-point channels (pressure/tilt/width/alpha/…), three-fidelity model (exact/native/raw), IR-JSON serialization.
  • Format registry + CLI: formats/__init__.py registry, inkterop convert / inkterop inspect.
  • Readers: reMarkable (ported onto the IR with zero regression — golden byte-for-byte drawing-op tests plus a 110/110 whole-library A/B check against the pre-IR renderer), xopp, IR-JSON, InkML, GoodNotes (ink + color only, experimental — pen-type field still unlocated, see docs/formats/goodnotes.md), Supernote (raster-first), Notability legacy (zip+plist), Notability modern .ntb (FlatBuffers op log decoded 2026-07-09, render-validated vs the app's own thumbnail; same encoding as the app's local-persistence note blobs — see docs/formats/notability.md).
  • Writers: PDF, SVG, InkML, xopp, IR-JSON — all validated = True per docs/validated-writes.md (open formats or our own; no native-app writers yet).
  • Golden fidelity harness: core/tests/test_golden_remarkable.py pins the validated renderer's drawing ops so the IR refactor can't silently regress fidelity.
  • RE toolkit: tools/re/{inventory,pbwire,applelz4,fbwalk}.py.
  • Docs: docs/ir.md, docs/formats/{remarkable,xopp,goodnotes, notability,supernote,inkml-mapping}.md, docs/reverse-engineering.md, docs/corpus-protocol.md, docs/validated-writes.md, all with confidence markers on unverified claims.
  • License: MIT.

M2 — complete in/out push ✅ LARGELY DONE (2026-07-09)

Landed on the note-apps workstream (18 readers / 12 writers total):

  • Native writers, all five (each validated = False behind --experimental per docs/validated-writes.md, awaiting app-open checks — samples staged in corpus/validate/): Saber, reMarkable .rm/.rmdoc (int-exact device-fixture round-trips), Supernote (raster RATTA_RLE, cross-validated via supernotelib), GoodNotes (protobuf/LZ4/tpl encoders), Notability .ntb (hand-rolled FlatBuffers builder).
  • New readers: Excalidraw (r/w), generic SVG (re-ingests our own SVG output — SVG is now two-way) + Stylus Labs Write, Wacom UIM 3.0/3.1 (oracle-validated vs Wacom's Apache library), MS-ISF (codec-exact vs Microsoft's WPF implementation), OneNote .one (full classic ONESTORE parse + undocumented ink hierarchy), Samsung .sdocx, Onyx Boox .note, PencilKit .pkdrawing (own novel RE, oracle-validated — the Apple Notes ink core), tldraw .tldr.
  • Per-format specs in docs/formats/ with confidence markers.

Still open in M2 scope: - Writer validation session (app-open checks → flip validated); Notability additionally gated on the red-ink color-byte-order corpus case; GoodNotes on Mac-import member-set iteration. - GoodNotes/Notability corpus deepening: pen-type/pressure fields (cases 05/07/14/16), curvesfractionalwidths (case 16). - Apple Notes NoteStore plumbing (TCC-gated) on top of the PencilKit decoder; rnote (needs drawn samples). - PDF fidelity (/BM /Darken exact blend; filled-outline variable-width strokes) and annotated-PDF base-page merge — carried over, unchanged.

Phase 2 — desktop app (sync control center + converter GUI) ✅ DONE (2026-07-10)

Shipped bigger than planned: a full Tauri 2 app (app/) instead of a menu-bar-only shell — one codebase gives the main window AND the tray item, and Windows/Linux builds fall out of the same code. ALL logic stays in core/: the app spawns inkterop daemon (stdio JSON-RPC sidecar; PyInstaller-frozen in release bundles, live uv tree in dev).

  • Core sync module (core/src/inkterop/sync/): multi-source (reMarkable cache, note-file folders, experimental GoodNotes/Notability container scanners), multi-sink (pdf/svg/png/inkz), per-doc rules (allow/block, rename, destination, format) in rules.toml, v2 state with v1 migration, watcher pid-lock. mirror/watch unchanged for launchd.
  • App: library browser (Finder-style columns / thumbnail grid / details list, first-page previews, folder+note sync toggles, output overrides), drag-and-drop converter over the full registry, activity log, settings, tray with sync controls, legacy-launchd migration.
  • Releases: .github/workflows/release.yml — macOS signed+notarized (once Apple secrets land — see app/RELEASING.md), Windows/Linux community builds; Homebrew cask recipe documented.
  • Website: website/ landing page at the Pages root, docs moved under /docs/.

Phase 3 — Import lane + iPad trial ◻ NOT STARTED

  • inkterop send <pdf> [--folder X] via ddvk/rmapi (mkdir+put), ALWAYS preceded by rmapi backup (official corruption warning).
  • inkterop send --editable via drawj2d 1.4+ (PDF/SVG → .rmdoc native editable ink, full Paper Pro color).
  • iPad app trial order (rationale in docs/research.md): 1. Saber (only open-format iPad app; verify writing feel), 2. Nebo (best feel; no longer one-way — .nebo BINK ink is now decoded, formats/nebo/),
  • OneNote (best automation: Graph API ?includeinkML=true). GoodNotes/Notability rejected — closed/uncracked (GoodNotes) or format being replaced mid-crack (Notability).
  • Stretch: Saber .sbn2 ↔ .rm bidirectional bridge (both formats open).

Phase 4 — Later

Windows/Linux shells; app distribution (notarization/updates); OneNote InkML puller; exact template art; publish for other reMarkable users.