transcribe-cpp is now the only on-device engine on every platform, so
the sherpa-onnx packages, their packaging (extraResources, the darwin
libonnxruntime prune, the native-variant assert and the Linux libstdc++
exemption), the sherpa recognizer in the speech process and the
per-model engine choice all go. Hotwords are written as plain terms.
* feat(speech): move the speech catalog to transcribe-cpp models
Parakeet v2/v3, Whisper tiny and SenseVoice now download their GGUF
builds, which run on transcribe-cpp everywhere, Intel Macs included (on
the CPU). A model still holding its old sherpa-onnx download shows
"Update to the faster version (<size>)"; updating replaces the old
files, and dictating before then says the model needs that update.
The five models with no GGUF build are retired and their users moved
over: the Japanese Parakeet, the Chinese-English Zipformer and the
Paraformer to SenseVoice, and the Korean and Chinese streaming
Zipformers to the new Nemotron 3.5 (28 languages). Nemotron's locale
codes (ko-KR) are matched from the ISO language setting.
* feat(speech): offer the speech model update once after the engine change
A card in the notification corner tells people whose selected on-device
model needs the new download, or whose retired model was replaced, and
starts the same download as Settings > Voice. After a replacement it
offers to delete the retired model's files.
* feat(speech): move retired Korean and Chinese models to SenseVoice and offer Whisper Large v3 Turbo
On transcribe-cpp 0.4.1 with Metal, SenseVoice beat Nemotron 3.5 on Korean
and Chinese FLEURS clips, so the retired Korean and Chinese streaming
Zipformers now move to SenseVoice. Whisper Large v3 Turbo (Q8_0, 886 MB)
joins the catalog as the most accurate choice for Japanese, Korean and
Chinese. Nemotron 3.5 stays, described by the languages it does best.
* feat(mobile): keep dictating after leaving the session, with a floating pill
* feat(mobile): translate the dictation pill and its toasts
* fix(mobile): type away words without claiming the terminal, and keep the pill above screens
* fix(mobile): float the pill above bottom bars, reopen the owner's own screen, and tell a refused start wherever you are
* fix(mobile): float the dictation pill above the source control bar
* fix(mobile): keep pill state out of the host context and reopen the starting screen first
* refactor(mobile): subscribe the pill only while it shows and trim the session dictation hook
* fix(mobile): key bar clearances by symbol
* feat(mobile): save the chat when the app leaves the foreground
A swipe-kill from the app switcher can end the app before the page is
hidden: iOS fires visibilitychange only at background, and Android Recents
leaves the activity resumed. The shell now sends an appInactive frame on
iOS inactive and Android window blur, to pages that declare they read it,
and the page saves any chat rows the store doesn't hold yet.
* fix(mobile): keep the leave callback out of render
* fix(mobile): notice a changed middle row before skipping a chat save
* perf(mobile): show accounts from the last snapshot
* fix(mobile): never count a stored accounts snapshot as live
* test(mobile): pin when accounts usage counts as live
* fix(mobile): date accounts reads by when they were sent and block switches until live
* fix(mobile): show accounts usage only from this screen's own read, drop replies from before a reconnect
* feat(settings): add a Use GPU switch for on-device dictation
On by default (Auto: the GPU when it works, the CPU when it fails).
Turning it off keeps the transcribe-cpp speech process on the CPU, for
machines whose GPU driver freezes or restarts them. The row shows when
the selected model runs on transcribe-cpp.
* fix(speech): reload the warm speech model when Use GPU changes
A loaded recognizer kept the GPU choice it started with, so turning Use GPU
off left the next dictations on the GPU until the model unloaded.
* fix(speech): reload the speech model after a dictation when Use GPU changed during it
A recognizer still in use when Use GPU changed stayed loaded afterwards with
the old choice. It now unloads as soon as its last dictation ends, and the
warm-up loads it again with the new choice.
The pill showed only the current partial and went back to Listening after
each final, so pause-committed phrases from transcribe-cpp never appeared.
The host's transcript fold moves to src/shared and the desktop pill uses it
too: committed text in the foreground, tentative text muted.
Add an OpenAI-compatible model whose base URL, model and optional key
are set up in a Voice settings dialog and sealed together in ~/.orca,
so a key is never sent to a host it wasn't saved for. It reaches
OpenAI, Groq, Mistral and local servers, and replaces the unused
userModels setting.
* feat(speech): run GGUF models on transcribe-cpp, falling back to the CPU
The speech host child can now load a second engine. A model whose files
are GGUF runs on transcribe-cpp (Metal on Apple silicon, Vulkan on
Windows and Linux, CPU otherwise); everything else stays on sherpa-onnx,
which remains the only engine on Intel Macs.
A GPU that fails to load, errors, stalls for 25 s or crashes the process
moves the dictation to a CPU-only process (TRANSCRIBE_BACKENDS=cpu) that
replays the audio the GPU had not decoded yet, so no words are lost or
repeated. A GPU that crashed or would not load is remembered per
transcribe-cpp version and OS release, and voice.speechGpu = 'off' keeps
it on the CPU.
Packaging copies transcribe-cpp, koffi and the target's native package
into resources/node_modules outside asar, fails when a target's package
is missing, and ships the MSVC runtime beside transcribe.dll on Windows.
No catalog model is GGUF yet, so nothing runs on transcribe-cpp until
the catalog change that follows.
* fix(speech): keep the speech queue running after a failed task, and drop audio frames after a discard
* fix(speech): keep the engine's file reads off the main thread's hot paths
The engine lookup and the GPU failure marker are read once per process
instead of on every dictation start and mobile catalog request, the marker
is written asynchronously, and the child reads the hotwords file with
fs/promises. The baseline moves the existing engine presence checks from
stt-worker-paths to the shared package table and adds the once-per-process
marker reads.
* fix(mobile): stop Review's branch compare re-reading back to back during git bursts
fetchHostQuery re-read a compare as soon as it landed marked by a git event, so a burst of gitStatusChanged events with reads slower than the event spacing kept the desktop running compares back to back, and kept going after Review closed. A re-read now goes at once only after a quiet spell (a write's own event), otherwise floorMs after the last read landed, and stops once Review's build is no longer current. A superseded build no longer evicts the entry a newer build is reading.
* fix(mobile): show each compare that lands during a git burst while Review reads it again
* fix(mobile): keep note edits when a later compare lands on a shown Review build
The phone filters in-progress VM checkouts only when it reads its host
workspace list, so a read made before a window's hide reached the main
process kept showing the checkout until the next unrelated change. Emit
worktreesChanged for each repo whose pending checkout appears or goes
away; phones already re-read their host workspace list on that event.
* fix(vm-recipes): hide an in-progress VM checkout in every window, the phone and the jump palette
A VM create's checkout was hidden only in the sidebar of the window that
started the create. Windows now publish their pending VM checkouts to the
main process, which shares every window's set with all windows and
filters the paired phone's host workspace list and workspace context.
The jump palette and the sidebar read the shared set, and a grouped VM
project no longer draws an empty-project placeholder while its create
is pending. A window's set is dropped when it closes, crashes or reloads.
* test(project-groups): reset leaked runtime environment revisions between files
* perf(mobile): open Home with last-known workspaces and counts
* fix(mobile): keep disconnected desktops on Home and judge agents by the client's connect time
* fix(mobile): store the seeded Home counts before removing the old snapshot
* fix(mobile): floor router reads after failures, catch up reads that span a stream reopen, seed the wide host list from Home
* fix(mobile): keep the old Home snapshot until the store can take it, and never re-date stored totals
* perf(mobile): key Home's counts in the host list module so the page boot never loads Home's code
* test(mobile): type the Home catalog adapter's modules with type imports
* perf(mobile): read Home's stored rows in one batch beside the host list, and keep mark-read a delta read
* perf(mobile): paint Home's desktop rows without waiting for the stored counts
* perf(mobile): restore Home's stored rows after the desktop list paints, from its host ids
* perf(mobile): read Home's counts on the connected edge itself, not a render later
* feat(mobile): keep commit, PR and review drafts across kills
* refactor(mobile): build commit, PR and review drafts on the shared draft fields, and clear only the text that was sent
* fix(mobile): keep a toggle made during Write with AI, the edit note while its sheet closes, and a newer PR edit after an earlier save
* test(mobile): wait for the saved review note row instead of sleeping
* perf(mobile): keep a hash of the PR text an edit was typed against, not a second copy
* feat(mobile): open chats on their last messages after a relaunch
The phone's store keeps each chat's newest rows (up to 512 KiB, 500 rows)
and its resume point, written when a turn settles, on hidden and when the
chat closes. A relaunched session opens on its stored chat tab, paints the
saved rows, and resumes from the stored cursor; the live read restates what
the agent is doing. Offline, the composer line says when the copy is from.
* test(mobile): keep the relaunch test typechecking
* i18n(mobile): translate the offline as-of line
* perf(mobile): encode chat copies in slices off the settle frame
* perf(mobile): keep three visited terminals mounted and release hidden WebGL
* fix(mobile): scope the mounted-terminal LRU to the terminal frame
* fix(mobile): keep chat's lease-only stream when a terminal pane unmounts
* fix(mobile): report a hidden terminal's cell box only once it is shown
* perf(mobile): keep hidden terminals' WebGL contexts instead of rebuilding them on every switch
* feat(vm-recipes): show VM lifecycle states in the sidebar, with Suspend/Resume VM
* fix(vm-recipes): Suspend VM suspends an idle parent with active children; dedupe repeat suspends
- With active children, Suspend VM sleeps this workspace too, so its VM
suspends once their sessions stop.
- A repeat Suspend VM while one is pending is dropped in the coordinator
instead of trusting a possibly stale suspended record.
- After a direct suspend, re-probe the host so its header reads Suspended.
* perf(mobile): back off spare page refills after an unclaimed drop or memory warning
* fix(mobile): skip a spare whose store read outlived an unclaimed drop
* fix(mobile): arm the spare refill timer on whole milliseconds
* fix(mobile): floor event-driven reads after failures and catch up reads that span a stream reopen
* test(mobile): pin the event router's read floor after a refused read in a recording
* feat(mobile): keep Tasks and new-workspace drafts across kills
Task comments (per item and project row), the New issue title and body,
and the New workspace note now live in device-store draft rows. They come
back after the form closes or the app is killed, and only into an empty
field. A comment's setter stays bound to its item, so a send that settles
after the user moved on clears the right row.
* fix(mobile): rebuild comment sends when the draft's item changes
* fix(mobile): key the workspace note by host and keep cleared fields cleared
* fix(mobile): clear only the note a finished create sent
* fix(mobile): count clearing an empty draft field as the user's choice
* perf(mobile): cap field drafts, read sheet drafts on open, retry a failed read once
* fix(mobile): retry a failed draft read after a pause and keep New issue text typed during a create
* fix(mobile): save draft text again once a failed store read can be retried
* fix(mobile): remove a stored draft when an unread field is cleared
* perf(github): use the desktop's PR fallback rule on the phone and forward refreshes
* fix(github): keep a merged fallback PR while the phone has no head
* feat(runtime): tell phones when relay hosts change
* fix(runtime): keep the relay host backstop live and watch every list input
* fix(mobile): hold the relay host feed only for desktops that announce changes
* perf(runtime): watch relay host inputs only while phones read the list
A recipe VM create registers the VM's project before its environment
checks run, so the VM's primary checkout showed as a normal workspace
next to the pending row while the checks ran or were paused on a
failure. Hide that checkout in both sidebar list models until the
create adopts it (or makes its worktree). A cancelled or failed create
keeps it hidden while its VM is cleaned up; a VM that could not be
cleaned up is shown again so it can be removed.
Codex reports both approved and denied automatic approval reviews on the
guardianWarning channel. Orca rendered every one as a yellow "safety review
warning", so routine approvals looked like problems. Approved reviews now
render as a neutral "approved this action" notice on desktop and mobile;
denials and timeouts keep the warning tone.
Co-authored-by: Sean876-alt <224890246+Sean876-alt@users.noreply.github.com>
Co-authored-by: Neil <neil@stably.ai>
The runtime service and failed-start cleanup now reach orca.yaml environment
recipes and plugin VM recipes through ShellRecipeDriver's recipe-shaped steps.
Same arguments, same runner, same results: records, failure copy and
diagnostics are unchanged.
* refactor: group agent launch, status, and messaging owners
* refactor: consolidate direct owner imports
* refactor: import pane ownership directly in status listeners
One lifecycle seam (create/resume/suspend/destroy/status + capability flags and a
catalogued error type) that every environment provider implements, a shared
vitest contract any driver runs from its own test, and ShellRecipeDriver, which
wraps today's orca.yaml / plugin VM recipe runner and passes the contract.
No production callers yet.
* test(e2e): give the provisioned-root VM a reachable git remote
VM creates now run the environment checks before creating the workspace
(#40152). The fixture VM's origin pointed at a GitHub URL it cannot push
to, so the git push check failed and the create paused on the build
sheet; the spec then read the VM's primary checkout before it was
adopted. Point the VM checkout at a local bare repo so the checks pass.
* test(e2e): wait for the provisioned root's adoption, not its listing
The VM's primary checkout appears under its branch name once its project
is set up, before the environment checks finish and adoption writes
ephemeralVmCheckoutMode. Poll for the adopted row instead.