Setting a channel makes electron-updater allow downgrades, and the channel
check let prerelease channels accept them, so a stable install on beta would
be asked to 'update' down to an older beta.
SQLite lower() only folds ASCII, so importing 'äpfel' next to an existing
'Äpfel' created a second local entry that later collided with the server's
unique index and silently overwrote one spelling. Import now compares with
the same JS identity key as add/update, including duplicates within the file.
Signing in as a different account (e.g. while the previous one was still
waiting on an offline restore) swapped the database under a live
recording, so its final save and audio upload went into the new account.
The sign-out capture shutdown now also runs whenever the open database
belongs to another user.
Checkpoints stay local until stop, so a pull during recording replaced the
saved transcript with the server's empty one; a crash then recovered the
meeting as an error with nothing in it. Status, transcript, end time and
duration of a locally recording meeting are now kept on pull.
The installer is unsigned, so latest.yml is now signed with an Ed25519 key and the
app verifies the downloaded installer against it before installing.
npm run release -- patch|minor bumps, gates, tags, builds, signs, publishes,
re-verifies the public feed and pushes; --resume continues after a failure.
The retired GitLab registry publisher is removed.
While suggestions are on, typos are no longer replaced silently. The popup shows
the fix next to the candidate and applies it only when the candidate is accepted.
The per-minute request cap rises from 6 to 20 so suggestions keep appearing after
accepting one and continuing to type.
The PC-wide slowdown was not a leak but model residency: boot warmup loaded
gemma4:e4b (~11.6 GB RAM incl. 5.9 GB pinned host buffer, no mmap on Windows
CUDA) and kept it for 30m, next to the large-v3 sidecar. Measured on the HEAD
build: free RAM 7.9 GB -> 32 MB within 20 s; after this change 10.8 GB with no
Ollama model loaded until typing starts.
- single LOCAL_LLM_KEEP_ALIVE (10m) for every local request; no boot warmup
- unload this app's resident Ollama models (keep_alive 0) on quit
- sidecar POST /unload for the aux slot; captions release it on stop
- partial dictation copies only the tail window, not the whole recording
- cap the voice-conversation listen buffer at 5 minutes
- mic test: generation guard so re-entry cannot orphan lease/timers
- startRealtime: single-flight and removeChannel to stop duplicate bindings
- meeting WAV save writes header and PCM without a concat copy
Docs: AI-01 policy note, GAP-PERF-01 with remaining work.
Typo auto-correct (INPUT-19)
- TypoCorrectionService + core typo-correction: 2-set Korean layout
mistypes both ways, always-wrong spellings, near-misses of the user's
frequent words, optional model check re-validated in core; applied by
backspace N + paste and cancelled on new keys or window change.
Prediction quality (INPUT-20/21)
- gemma models are sent raw with the user's unfinished sentence pre-filled
into the model turn, so the model truly continues the text (mid-word
joins and spacing exact). Other models keep the instruction prompt.
- Context: text before the caret (opening lines and a fact list of names,
dates, amounts outside the 400-char window), messages just sent in the
same chat window (memory only, 30 min), personal-graph follow-ups and
IDF-ranked related sentences with Korean stemming.
- Guards: first sentence only, restated-prefix stripping (also fixes the
instruction path), repeat drop, no invented dates/names/numbers
(one retry), plain -다 register, no emoji unless the user uses them,
search-box keywords. Fill loop mixes instruction prompts for variety.
- Opt-in real-model probe: tests/integration/suggestion-quality.ollama.test.ts
Suggestions now actually show up
- Terminal (Claude Code): caret trust from the input line, not the whole
screen (status line below the box blocked every keystroke); Korean
composition counts as typing at the end; a grey placeholder filling the
empty box at once is not user text.
- Chrome: UIA focus flicker while typing no longer resets "edited".
- KakaoTalk/RichEdit: trailing paragraph mark ignored for edit position
and stripped so accepted text keeps its separating space.
- While the overlay is visible the suggestion shortcuts are captured so
Ctrl+Alt+Enter no longer also types a newline; released when hidden.
- nut-js preloaded after boot (first accept waited 4.8 s).
CI run 121 failed at "버전·계약 정본 대조": earlier refactors moved update
gating from UpdateService into update-policy.ts (evaluateUpdateOffer ties
decideUpdate + isWithinRollout), the differential-download switch into
update-adapters.ts, and the Forgejo re-release guard into
lib/immutable-package-guard.mjs (sha256 comparison, abort on conflict).
The checks still looked for the old markers in the old files. They now
verify the same guarantees where the code lives, and the self-test's
negative case targets assertNoImmutableConflicts.
Also clears the lint gate: a control-character regex in Keycap (range now
starts at U+0020) and unused helpers in two red-team tests.
Wave 3 of the 2026-09-28 design overhaul.
- Esc (no modifiers) cancels an in-progress dictation, recording or
processing: VoiceModeService.requestEscape() queues the dead 'escape'
action. Clicking the capsule cancels only while recording, so a stray
click during processing no longer throws the text away.
- Errors show in the user's language: bootstrap now passes error.code to
the tip (the translated branch never ran), and one core classifier
(voice-error-message) feeds both the tip and the main-window alert.
- Capsule: bars stay flat without input (no fake "alive" wobble before the
mic opens), an indeterminate bar replaces the made-up percentage, the
partial transcript stays visible while processing, an Esc hint, and a
brief "Inserted" state so success is visible.
- All six popups share popups/_shared (Pretendard, one panel radius,
selection, kbd hints, reduced motion). History/command popups get
titles, listbox roles and window sizes from popup-list-geometry (empty
states were clipped); the command popup's "0" key is registered;
result popup explains why it appeared and can be closed; captions keep
a theme-independent dark backing on purpose.
- Popup themes follow the OS when set to system and the document lang
follows the app language.
- Site hero demo mirrors the new capsule (done state, indeterminate bar,
Esc hint; 10 locales).
Batch of red-team r3 fixes that were in the working tree before the
2026-09-28 design overhaul, committed as one unit with their tests.
- desktop main: STT timeouts and sidecar, voice recording store, sync
(credentials, audio, knowledge reindex, push gates), runtime
provisioner, update policy, AltGr keybindings, voice-command policy,
dictionary file codec/limits, meeting transcript condensing and a
local recording ledger so interrupted-session recovery only closes
meetings this device recorded (a phone's live meeting is left alone).
- mobile: login CSRF via implicit token callbacks rejected, account
deletion/retention, durable queue retention, knowledge realtime
without unfiltered DELETE, meeting re-record failure paths, cloud STT
client, preferences store/resync.
- core: text chunking splits long unbroken transcripts to fit, template
field policy, dictionary limits, meeting markdown inline handling.
- server: payple webhook policy and cancellation order scope, meeting
document generation quota, team RPC null-role guard, unified LLM
quota in-flight accounting, knowledge chunk vector index, meeting
re-record failure paths (migrations 20260929*).
- ci: portable/runtime feed gates, update-policy schema, Forgejo file
delete and alias planning.
Four older tests are updated to the new contracts rather than the old
behavior: token-pair auth callbacks are rejected, knowledge realtime no
longer subscribes to DELETE, long transcript lines are split, and
meeting recovery requires the local recording ledger for empty rows.
With a 100 ms trigger left in the installed config, generation started in the
gaps between keystrokes; the next key ended the session silently, and that
session still counted against the 5 s minimum interval, so the moment the
user actually paused nothing came (rate-limited). The log showed a
generate-then-vanish cycle every 5-6 s.
The trigger delay now has a 500 ms floor and settings revision 6 resets a
stored value below it to 600 ms. A session ended because the user kept typing
no longer blocks the next request by the minimum interval (per-minute and
daily caps still apply), and that dismissal is logged.
The phone draws a meeting from its transcript segments before the edited
transcript, so desktop edits, auto-polish and diarization never showed there.
Every desktop transcript change now rebuilds the meeting's segments from its
[MM:SS] [speaker] lines and trims the rest; the line parser moves to
@d3ro/core/meeting-transcript and the meeting view uses it too.
Prompt edits of the four desktop presets that exist on the phone update the
server preset row (a reset restores its default; {{targetLanguage}} is sent
as English, the only target on both sides), and edits made on another desktop
come back. The free-prompt preset has no phone counterpart and stays local.
Knowledge documents travel as source-text chunks; each surface embeds them
with its own model, the server index is requested through embed-chunks, and
documents from the phone are stored without a file and indexed from their
chunks. Chunk text is now kept when local embedding fails, so reindexing no
longer needs the original file.
Recordings upload to the mobile storage contract (audio bucket under the
user's folder plus an audio_files row, 50 MiB cap, a Settings > Cloud
toggle) and are removed with their record. The history card gains a play
button that uses the local file or, for phone recordings, a signed URL.
Language (ko/en), system/light/dark theme, auto-polish and the active user
command follow the phone's user_settings with its revision rule; changes that
arrive from the phone reach the open window.
Rewrites the desktop mirror as services/sync/SyncEngine: a persistent
outbox, per-account server-clock keyset cursors with paging, pulls that never
overwrite unsent local edits, deletions both ways through sync_tombstones and
per-row failure isolation. It now covers history titles and favorites,
dictionary, every meeting's memos and documents, memo tags, user commands and
dictation/meeting templates, and registers the desktop as a device that the
phone can disconnect.
Fixes shipped defects: the first pull after sign-in fetched nothing, only
the first meeting's children were pushed, team meetings leaked into the
personal database and lost team_id on re-push, and Realtime never connected
because Electron's Node 20 has no global WebSocket (ws is now the transport).
Anonymous local-mode records are imported into the first account that signs
in. The settings sync section is translated and shows pending/rejected
changes; synced screens reload on app:dataChanged.
With the embedding server unavailable every chunk failed, yet the document
was stored as indexed=true with 0 chunks, so the knowledge base listed it as
searchable while queries could never match it. The red use-case test caught
this; it had been written off as an environment failure.
Now a run with zero embedded chunks leaves indexed=false and throws
RAGEmbeddingFailed (surfaced by reindex, logged by addDocument).
Tests that only hold on the Windows developer machine now declare it: the
bundled SoX binary and PowerShell device discovery run on win32 only, and
the sidecar venv test runs only when sidecar/.venv exists. The Linux Forgejo
runner skips them instead of failing.
The speech engine now keeps an auxiliary model next to the dictation model
and transcribes with whichever the request names, reloading it once if the
engine restarted. Settings > STT gains a live-caption model so captions can
run on large-v3-turbo while dictation keeps its own model. The runtime
minimum rises to 1.7.0 because older engines would silently ignore the
model choice.
Suggestion paging moves to Up/Down: the page follows the selection and the
last item waits while more candidates are being generated. The Left/Right
page shortcuts are removed; they did nothing until a page had filled and
clash with Intel's display-rotation hotkeys.
Replaces the fixed six-second batches with a streaming track per audio
source: the uncommitted audio is re-recognised every second and sent as a
partial with its agreed (stable) prefix, a short pause finalises the line,
and long unbroken speech is committed at Whisper segment boundaries. Idle
audio is trimmed so silence cannot produce invented sentences.
Finished lines are corrected by the local model against the previous lines
and replaced in place; edits that change too much are rejected. The
behaviour can be switched off in Settings.
Next-sentence suggestions now arrive one at a time up to twelve, shown three
per page with Ctrl+Alt+Up/Down to move, Left/Right to page, Enter to accept
and Esc to close; old default bindings migrate and the panel guide follows the
live bindings. The overlay is redesigned, stays put while candidates stream
and sits outside the input box when no caret is reported.
The personal phrase memory stops learning from terminals, code editors and
the coding-agent hub, ignores symbol-heavy lines and empty-field placeholders,
and prunes existing entries that break those rules.
Fixes suggestion keys starting dictation, installs stuck on a pre-1.5.0
speech engine without the focus endpoint, Ollama runner windows flashing
while typing, the speech engine starting twice, and cold-model timeouts.
Live captions can be dragged to a remembered position and show a waiting
notice until the first line arrives.
Bumps the product version to 1.6.0 (Android/iOS build 1060000).
Adds next-sentence suggestions while typing, weekly input insights and a
personal phrase memory to the desktop app, and fixes custom instructions so
they process the text instead of inserting the instruction's own wording.
Local model requests are now bounded and individually cancellable.
Bumps the product version to 1.5.0 (Android/iOS build 1050000), refreshes the
landing and web download links, and records the new INPUT feature rows and the
open verification gaps in the infrastructure map.
Running a custom instruction (translate, summarise, rewrite, explain code,
free prompt) inserted the instruction's own wording instead of the result.
Two faults stacked:
The instruction was passed as the text to process, leaving the system-prompt
argument empty. `BASE_SYSTEM_PROMPTS` has no `custom` key, so resolution fell
back to `refine` without saying so, and the model dutifully polished the
instruction it had been handed. The transcript never reached it.
And only `{{text}}` was substituted, which none of the five built-in
instructions use — they carry `{{targetLanguage}}`, `{{userPrompt}}`, or no
placeholder at all. The substitution was a no-op from the day it was written:
the presets landed ten hours before the code that expected them.
- Instruction prompts now go to the system-prompt argument and the transcript
to the text argument. Instructions that spell out `{{text}}` keep their old
meaning, so hand-written ones still work.
- `renderInstructionPrompt` resolves `{{text}}`, `{{userPrompt}}` and
`{{targetLanguage}}` in one place, and warns by name when a placeholder is
left standing rather than letting it reach the model.
- `resolveSystemPrompt` no longer drops silently to `refine` for `custom`.
- Voice shortcuts no longer die at the `defaultLLMAction === 'none'` gate; an
explicitly named instruction outranks the default. Without one, `none` still
passes the transcript through untouched.
- `translate` receives its target language instead of relying on a default two
call frames away. It is still always English — `AppConfig` has no key for it,
and neither `language` (UI locale) nor `sttLanguage` (source language) can
stand in. Choosing a target language needs a setting and is not in this fix.
- Chains ran instructions with placeholders intact; they share the same
resolution now.
- The command screen's pipeline bench called `llm.generate`, which preload does
not expose, so every run threw and the catch showed the input back as if it
had succeeded. It uses `llm.process` now, over the same path production
takes, and a failure reads as a failure.
Present since the feature shipped: the custom-instruction path has never
worked. Plain actions (refine, summarise, grammar, expand) were unaffected and
are now covered by tests so they stay that way.
Shortcuts were defined in four places that drifted apart: per-action IPC channel
pairs, a hand-written VK table in the service, a second one in the renderer, and
three copies of the keycap styling. Adding an action meant editing all of them,
so two shortcuts stayed hardcoded in bootstrap and one had no settings entry at
all.
packages/core/src/keybinding.ts is now the single source for the binding type,
the selectable key catalog, the action catalog, normalization, validation,
conflict detection, display labels, search and deserialization. Main, preload
and renderer all read from it; nothing redefines keys or rules locally.
- Each action holds a list of bindings instead of one. AppConfig's four
*Shortcut fields collapse into a single keyBindings map, migrated on launch.
- Mouse buttons can be bound. Left click is refused, right/middle need a
modifier, side buttons are free. uiohook cannot swallow events, so the
original click still fires and the UI says so.
- Keys can be picked from a grouped dropdown with a search box, not only by
recording a keypress.
- HOTKEY's 14 channels become KEYBINDING's 9, taking the action as a parameter,
so actions no longer multiply channels. The history and command popups moved
out of bootstrap into ordinary actions.
- displayLabel is gone; labels derive from the binding and follow the app
language and platform.
Fixes found on the way:
- Double-press hands-free was unreachable: lookup returned only the first
matching action, and dictation shares its default binding.
- Reserved-combination checks compared joined key names, so a different modifier
order let Ctrl+C through.
- Disabling shortcuts released every global registration in the process,
including the popup ones, and never restored them.
- Enabling shortcuts after starting disabled left nothing registered.
- The dashboard stored the caption event payload instead of the state in it.
Local dictation had never produced a transcript on an installed build. The
engine itself was healthy; every connection to it was broken.
Installed builds shipped no speech engine at all: the packaging config had no
entry for the faster-whisper sidecar and no pipeline step built one, so the app
always fell back to a system Python without the runtime. Development was broken
too, because the sidecar and SoX paths were resolved against the Vite output
directory instead of the app root, which also meant recording failed with a SoX
ENOENT. On hosts where localhost resolves only to IPv6, every local request was
refused outright, which silently disabled both local transcription and the local
LLM.
The sidecar is now built and bundled (including the Silero VAD data it needs),
gated by a packaging check that fails when the engine or its data is missing.
Paths are discovered from the app root and fail loudly when the engine is
absent. Local engine URLs are normalized to the IPv4 loopback, decoding is tuned
so repeated hallucinations cannot compound (the same transcript now takes about
a fifth of the time), the engine is warmed up at startup, and holding the hotkey
now shows the text forming live in the recording tip.
Several desktop paths quietly substituted defaults or partial results: a
config write could fall back to a throwaway in-memory store, speech provider
errors were absorbed into empty transcriptions, and meeting exports built
file names from raw titles.
Writes now fail explicitly when the store is unavailable, provider and model
failures reach the UI as errors, and export names pass through one
sanitizer. Settings, license, ad, and support surfaces use the shared theme
tokens, unused hotkey helpers are gone, and the package gains strict
node/renderer typecheck configs plus red-team e2e scenarios for these flows.