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.
A candidate list was cleared about two seconds after it appeared: once typing
paused, the not-typing rule dismissed the visible overlay, so a click or
Ctrl+Alt+Enter found nothing to insert. The rule now only stops new
requests; a visible overlay stays until it is accepted, dismissed or goes
stale.
Terminals were excluded from suggestions because the whole screen buffer
reads as the input and the last line is usually a status bar. The input line
is now extracted (Claude Code/Codex >, starship, PowerShell, cmd and POSIX
prompts, wrapped continuation lines) and used as the prefix; no prompt means
no suggestion. Terminals stay out of phrase learning. Accepting with nothing
shown and successful inserts are now logged.
A five-second phone recording took over a minute: stt-proxy waited up to
60 s for the self-hosted gateway, whose GPU endpoint was off and whose NAS CPU
Whisper needs 30-90 s per clip. With a direct provider configured the gateway
now gets 5 s plus the clip length (30 s cap).
The direct OpenAI fallback never produced a result. The production key held
characters that are not valid in an HTTP header, so every request threw while
being built; provider keys are now stripped of BOM/zero-width characters and a
still-invalid key counts as not configured. whisper-1 verbose_json reports the
language by name, which the result contract rejected; names now map to codes.
Fail-closed responses list each provider's failure (status or error class,
no secrets) so an outage can be diagnosed without log access.
Desktop transcript edits, auto-polish and speaker labels now reach the phone,
which draws meetings from transcript segments, and prompt edits of the four
shared preset commands are used by the phone's commands.
Bumps the product version to 1.9.0 (Android/iOS build 1090000).
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.
Preset custom_instructions rows are not client-writable, yet the phone runs
every command with the row's prompt. sync_set_builtin_instruction_prompt_v1
changes only the prompt of the caller's preset (NULL restores the default),
sync_list_builtin_instructions_v1 reports whether each prompt is the default,
and the defaults live in one function.
Knowledge documents move between desktop, phone and web as source text and
are indexed on each device; desktop recordings upload to the shared storage
and any recording plays from the history card; language, theme, auto-polish
and the active command follow the phone. Document text survives a missing
embedding model so it can be indexed later.
Bumps the product version to 1.8.0 (Android/iOS build 1080000).
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.
knowledge_documents deletions now leave a sync tombstone and have a cursor
index and Realtime, so the desktop mirror can follow them. pg_cron runs
prune_sync_tombstones_v1 every day at 03:17 UTC with the 180-day window.
Signing in now keeps history, dictionary, meetings with memos and documents,
memo tags, user commands and templates the same on desktop, phone and web,
with offline changes queued, deletions shared and phone edits arriving live.
The desktop registers in the phone's device list and signs out when
disconnected there; local-mode records move into the first account.
Live captions stream about a second behind speech, finish lines on pauses,
are polished in context by the local model and can use their own speech
model (the speech engine is downloaded again once, minimum 1.7.0).
Fixes suggestion paste on Ctrl+Alt+Enter, click acceptance, typing detection
in growing chat boxes, one-way cloud sync and failing account exports.
Bumps the product version to 1.7.0 (Android/iOS build 1070000).
normalizeRegisteredDevice accepted only android/ios, so one desktop row made
the whole Devices screen fail. Accept the server platform list and show the
platform on each card.
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.
sync_tombstones logs every owner-scoped delete on the synced tables so an
offline device can apply it later; filtered Realtime channels never deliver
DELETE, so the tombstone INSERT is also the live deletion signal.
history.revision now follows content changes made without a revision (the
desktop mirror), meeting_memos gains updated_at, and cursor indexes back
keyset pulls. Templates get client-keyed upsert/delete wrappers over the
revision-checked mobile RPCs.
export_account_portability serialised whole rows while the v1 archive is an
exact key set; columns added later made every account with meetings fail
export. Each dataset is projected back onto its v1 keys.
Crawlers received an empty client-rendered shell (107 characters of text);
most AI crawlers do not run JavaScript, so the product was invisible to them.
- Build renders each locale to static HTML (/ for Korean, /en/ … /vi/) with
src/entry-server.tsx + scripts/prerender.mjs; the browser hydrates the same
locale. The language menu links to those pages instead of switching in place.
- Per-locale head: title, description, canonical, hreflang (+ x-default /en/),
Open Graph and X cards with a per-locale 1200x630 image.
- JSON-LD graph: Organization, WebSite, WebPage, SoftwareApplication (KRW
offers, version, release date, download URL from the canonical sources),
HowTo and FAQPage. No invented ratings.
- robots.txt (search, ai-input and ai-train allowed; /app, /api, invites
excluded), sitemap.xml with language alternates, llms.txt and llms-full.txt
generated from the same translations and canonical values.
- IndexNow key and a post-deploy ping (Bing, Naver, Yandex).
- A factual "at a glance" section and two FAQ answers (recognised languages,
where data is stored) in all 10 languages.
- Icons, apple-touch-icon and web manifest; asset base is now absolute so
sub-path pages load the same bundle.
Verified: 3.6k-7.7k characters of text per page, no hydration warnings,
no overflow at 320/1440 in six locales, axe 0, Lighthouse mobile SEO,
accessibility and best practices 100.
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 first real Forgejo CI run failed because the linux-builder runner ships
Node 18.19: scripts using import.meta.dirname threw, and its older npm
reported the mobile lock as out of sync with the file: dependency on
packages/core (the lock is in sync under Node 24's npm).
bootstrap-linux-toolchain.sh gains a checksum-verified `node` mode
(24.19.0, the .nvmrc version); ci.yml (quality, build-validation,
mobile-quality) and deploy-site.yml add it to GITHUB_PATH before npm runs.
