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.
* feat(dictation): dictate into the focused field of a page in the browser pane
* test(dictation): cover dictation into a focused browser page
* fix(dictation): type page dictation through the frame that owns the field
* fix(dictation): space page rich-text dictation within its block and offer editors the beforeinput
* test(dictation): simplify the page insertion test mock
* fix(dictation): copy page dictation to the clipboard when an editor blocks the input
* feat(plugins): add workspace.rename and project names in workspace.list
Plugins that name workspaces had to read Orca's private runtime token and call
worktree.set over the runtime socket, bypassing capability consent. workspace.rename
(workspace:write, worker only, host API 1.5) does the same update as a sidebar
rename. workspace.list rows gain the project name so plugins can prefix names
without reading the runtime's repo list.
* fix(plugins): harden workspace.rename against twin folders and hidden characters
Folder renames require a unique folder id, so twins on two hosts are not both renamed,
and fail when no folder has the id. Names refuse C1, bidi and zero-width characters,
which could make a label read as another workspace's name.
* fix(plugins): workspace.rename matches a sidebar rename for folders and reports kept names
A folder rename also cancels a pending first-agent-message rename, so the automatic
title cannot replace the plugin's name. A rename Orca does not apply (during a WSL
move) is a conflict instead of a silent success. workspace.list's project is optional,
so mocks written for host API 1.4 still validate. More invisible characters are refused.
* refactor: move atomic publication to its filesystem owner
* refactor: group shared browser definitions by feature
* refactor: bind proxy process and session contracts to defining owners
* refactor: consolidate direct session contract type imports
* refactor: retain atomic publication test timing under moved owner
* fix(benchmark): resolve historical tunnel framing source before loading
* fix(speech): send cloud dictation through the proxy set in Settings
* fix(speech): keep OpenAI cookies out of the desktop session
* refactor(speech): send transcription and the key check through one request helper
* fix(worktrees): offer nested-worktree delete for remote runtime hosts
A remote host's 'contains another registered worktree' refusal showed only
the raw error. Add worktree.previewNestedRemoval and an approved-plan
worktree.rm path on the runtime, and route the toast's nested-delete dialog
to the workspace's own host.
* fix(worktrees): re-check checkouts between remote nested deletes; scale reply timeout
* chore: compose nested worktree ids; record improved ratchets
---------
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
* feat(vm-recipes): add Set up with an agent to the Cloud VMs settings
Starts the default agent in the project's workspace with a prompt that loads
the per-workspace environment skill, loops on the recipe and environment
doctors, and ends with a commit and PR. The launcher also takes failure
context (redacted log tail) for callers such as a failed VM create.
* refactor: split recipe log redaction into a browser-safe module
* fix(vm-recipes): bound and redact failure text in the setup-agent prompt
* feat(mobile): cancel dictation in place and show which device is transcribing
* fix(mobile): give each composer its own dictation cancel target
* fix(mobile): keep the dictating composer row on one line
* refactor(mobile): share the dictation mic face and measure the cancel target per platform
* fix(orchestration): keep a paired server's worker list when one Dispatch is from an earlier pairing
The server's fleet snapshot refused the whole batch when one requested
Dispatch was not this home's (for example, one started before the home
re-paired), so the home showed every worker on that server as
host_unavailable. Now only that Dispatch reads as unverifiable.
* test(orchestration): cover a live Dispatch held for another home in the fleet snapshot
* test(orchestration): assert the other home's fleet item without a cast
* feat(cli): add orca vm env doctor to check a workspace environment
Checks required CLIs (git plus what the repo manifests imply), agent CLIs
and their sign-in, Git fetch and a dry-run push, the orca.yaml setup script
(optionally re-running it with Orca's runner), and free disk. Pass/warn/fail
with --json; checks live in src/main/environment-doctor for runtime reuse.
* fix(env-doctor): ship the doctor modules in the packaged CLI and share the check formatter
* fix(env-doctor): run Git probes in the C locale so folder detection is locale-independent
One terminal claim owner now covers local Dispatches and attachments a paired
coordinator starts here, so either kind refuses the other with terminal_busy
before readiness waits or briefs.
* fix(claude): name Orca as the client of its Claude chats
Orca's Claude chats reported the SDK's default entrypoint (sdk-ts), or the
value of a Claude session Orca was launched from. Set CLAUDE_CODE_ENTRYPOINT
to orca for every structured Claude launch; a value in Orca's launch
environment still wins.
* fix(claude): keep a differently cased Windows entrypoint override
Windows env names are case-insensitive, so a launch value spelled
claude_code_entrypoint lost to the uppercase Orca default when Node
deduplicated the keys.
* fix(claude): let a launch value replace a differently cased inherited one on Windows
* fix(claude): store a Windows entrypoint override under the name the SDK reads
worker-start --on <server> --worktree new-child now tells the server that
orchestration picked the parent, so the server records the child's lineage as
orchestration, as a local child is, instead of cli. Older coordinators omit the
field and keep cli. The guide now shows new-child first and new-top-level as
the no-parent choice.
* perf(ports): index workspace ownership once per listener inventory
* perf(ports): stop command matching at impossible path prefixes
* Avoid unused port ownership index construction
* Match command ownership with bounded prefix transitions
* Fix the speed button's unknown, Flex and composer Escape cases
Follow-up to #40052 from its review:
- An unknown speed names no tier instead of claiming Standard.
- Escape in the speed list inside the chat options popover closes only the
list, not the whole popover.
- Only Codex's Fast tier gets the on/off chip, bolt and pressed fill; Flex
and other tiers are picked from the list. The Fast test is shared with
the main process.
- The list trigger takes the chip's pressed fill from the shared variant,
and the list has an accessible name.
* Show every agent's speed as the same Speed button in every picker
Claude's and Cursor's Fast mode now use the ⚡ Fast button under a Speed
label in the chat options popover and the select-mode pickers too, instead
of a Fast mode switch, matching Codex: one button for two modes, a list for
more.
* wip: error pill
* wip: error pill tests and strings
* wip: start failures with audio are model-load failures
* wip: tests
* fix(dictation): fresh pill per recording, keep the recording when a retry fails
* fix(dictation): retry a recording cut at the cloud cap; leave Esc alone on the error pill
* refactor(dictation): share the cloud cap and the pill's close button; fetch the catalog once
* feat(dictation): a retry types only the words after those that already landed
* refactor: separate execution route and child exit owners
* refactor: move browser data into defining feature owners
* refactor: give runtime RPC entry its shutdown implementation
* refactor: use defining process types and colocate host telemetry test
* docs(process-host): point piped child contract to Node owner
* docs(process-host): clarify both piped contract owners
* fix(remote-runtime): detect a runtime that stays connected but stops answering
The shared control socket only checked ws pings, which a relay hop or a runtime
whose request path is stuck still answers, so a host could read as connected for
hours with no updates (#39785). Ready sockets now send a cheap status probe after
an idle interval; a missed deadline retires the socket so reconnect backoff runs,
and the mark stays in diagnostics until the runtime answers again. Asleep and
manually disconnected hosts are never probed.
orcad also starts a worker-thread watchdog that logs a blocked event loop and,
when run under a service manager, exits after five minutes so it is restarted
(#34410).
* fix(remote-runtime): clear the unresponsive mark on any post-ready frame and ignore late deadlines
Review follow-ups: a reconnected socket that streams frames proves the runtime
answers, so it clears the mark and lets backoff reset; the probe deadline now
sits above the ws liveness window so a dropped link is not blamed on the runtime;
a deadline that fires late after sleep is not counted; integration tests no longer
depend on wall-clock timing.
* fix(remote-runtime): gate the heartbeat on the shared retry hold and isolate the wedged-runtime test
* test(orcad): wait for stdio close before reading the watchdog child's stderr
* feat(setup): slim the setup-script prompt and make long imported scripts collapsible
* fix(setup): require the full imported script on screen before saving
* style(setup): use the ghost button colour for the script toggle
* style(setup): keep the source path readable in the narrow prompt card
* feat(setup): title imports, restore the package-manager card, keep the review dialog inside the viewport
* fix(setup): count every script line, keep the e2e import spec in step with the new copy
* feat(dictation): fixed-size dictation pill with cancel, timer and finish
* fix(dictation): abort a cloud upload cancelled while finishing; time recorded audio
* refactor(dictation): reuse Esc/IME helpers, keep the pane on the target, isolate meter renders
* fix(dictation): pause the pill timer while hidden; compile under React Compiler
* feat(dictation): remove filler words from text the phone recognizes itself
Phones that dictate with their own recognizer (Automatic or This phone)
never send text through the computer, so its filler pass never saw it.
The page's on-device speech adapter now runs the same shared text
pipeline on the phone's live and final text, in the language of the
locale the recognizer started in, and honours the computer's Remove
Filler Words switch, which the host now reports in its dictation setup.
An older host that doesn't report it leaves the switch on, its default.
* test(mobile): expect both commas to go around a mid-sentence filler on the phone too
The active terminal refits a frame after the browser splits mount and can
refocus xterm after the spec focused the Reload button, so Ctrl+C reached
the shell instead of arming grab. Refocus until the toolbar keeps focus.
Co-authored-by: OrcaWin <293788423+OrcaWin@users.noreply.github.com>
OpenAI shuts gpt-4o-transcribe and gpt-4o-mini-transcribe down on
2027-02-26. Offer GPT Transcribe instead, retire both old entries so a
saved choice loads as the new model, and send the language hint as
languages[], which gpt-transcribe takes in place of language.
"I think we should, uh, ship it" became "I think we should, ship it":
the filler's own comma went with it, but the one before it stayed. When a
pause follows the filler too, the comma before it now goes as well.
- Browser run Show button: keep the remembered page in state instead of reading a module Map during render, so a run keeps the page it drove.
- Linear description toolbar: subscribe to active formats with useEditorState; the editor mutates in place.
- Usage tooltip: recompute the 'Updated' and reset-credit expiry labels on each countdown clock tick.
* feat(cli): create workspaces on environment recipes
`orca worktree create --recipe <id>` creates the workspace on a new
environment from one of the repo's environment recipes, and
`orca vm recipe list` shows which recipes it accepts.
The desktop app on the provisioning computer runs the same main-side
provisioning the composer uses (write-ahead record, per-step timeouts,
cancellation), registers the project on the environment's Orca server,
then creates (orca-worktree) or adopts (provisioned-root) the
workspace there. Named progress steps ride on long-poll keepalive
frames, so the CLI prints them to stderr; older CLIs ignore them. A
failure after the VM exists destroys it through the shared cleanup and
returns recipe_workspace_create_failed with the failed step, the
recipe failure code and whether the destroy finished. Ctrl+C aborts
the request and destroys a started environment.
Paired devices cannot start a recipe create. Hosts advertise
worktree.recipe-create.v1 only when the desktop wires the recipe
service, so orcad and older hosts are refused before anything runs.
Based on #39228 by jutaz, a port of #28439.
Co-authored-by: Justas Brazauskas <brazauskasjustas@gmail.com>
* fix(cli): report the VM outcome when connecting fails; record no cross-host parent
* fix(vm-recipes): destroy the VM when recording its environment fails; reset exitCode in CLI tests
* perf(types): keep recipe orchestration out of the runtime's type graph
---------
Co-authored-by: Justas Brazauskas <brazauskasjustas@gmail.com>
* feat(dictation): add an Auto shortcut mode where holding records until release and a tap toggles
* fix(dictation): apply the terminal shortcut policy and the shortcut recorder to held dictation presses
* test(dictation): drop casts from the shortcut mode tests
* fix(dictation): time Auto holds by the key, not a Cmd release macOS reports alone
* fix(dictation): start an Auto hold pressed while the last dictation finishes
* refactor(dictation): simplify Auto hold timing and reuse the shared IME check
* wip
* rm execessive tooltip
* Clarify deletion options and improve resource popover layout
- Show the scope of persistent deletion choices directly in the menu
- Keep resource details accessible in short windows
- Style disabled dropdown actions consistently
* Clarify detached commits and deletion guidance
* Preserve translated tooltip guidance
* perf(sqlite): refresh the legacy profile copy hourly and recover crashed row caches
- The legacy whole-document refresh rewrote every changed split domain each 5 minutes
(~4-8 MB per cycle, ~200 MB/hour of WAL plus checkpoint on an active profile). Writer
start and graceful close already bring it current, so refresh at most hourly in between.
- Cap the profile-state WAL at 8 MB; without journal_size_limit it stayed at the size of
the largest transaction ever written.
- Row caches: a read-only startup read cannot roll back a hot journal (or recover a WAL
without -shm), so retry once on a writable connection instead of disabling the store
on every launch.
- Row caches: drop the saved baseline when a commit throws, since COMMIT may report an
error after the rows landed and our own commit does not move data_version.
* test(sqlite): accept Windows exit status for the crashed-writer fixture; document recovery cost
* docs(sqlite): state the crash-recovery cost without overstating its bound
* fix(sqlite): rebuild a row cache a crash left needing recovery instead of replaying it on main
* revert(sqlite): keep crash-recovery failures non-destructive; drop the rebuild fallback
* refactor: use concrete browser and input owners
* refactor: bind remote runtime callers to concrete transport owners
* Use checked subscription mock argument types
* Type browser lease mock with its subscription contract
* refactor: bind sleeping VM status probe test to socket owner
* feat(setup): import setup scripts from more workspace config formats
* fix(setup): isolate imported Copilot steps and harden config parsing
* chore(setup): drop type assertions flagged by the quality gate
* fix(setup): keep the orcad bundle ASCII and skip unrewritable dev container variables
* fix(runtime): a phone viewing a terminal on a headless runtime no longer resizes it for everyone
On orca serve / orcad, a phone that only opens a terminal now views it at the
current owner's size. It takes the size when it types or when the user taps
the phone-mode key; the desktop's take-back reclaims it. Desktop hosts keep
fitting phones on open.
* fix(runtime): keep phone claims across leave grace and held phone fit; pin the headless default for the process
A Codex chat launched codex app-server on the account home saved when the
chat was created, and nothing compared it with the selection, so after a
switch every message kept going to the old account. A chat that had hit
its usage limit kept failing with "Codex hit its usage limit" even though
the new account had quota.
Codex chats now follow the selection the way Claude chats do: the launch
prepares the selected account's home, the started process reports that
home, and the host replaces an idle chat whose account no longer matches
before its next message. Codex resumes a thread only from a rollout in its
own CODEX_HOME, so the launch links the thread's rollout into the new home
first. The chat's usage card reads the account its next turn runs on.
* fix(native-chat): clear /compact from the composer once it is sent
Conversation commands kept their draft until the host replied, while
messages release it as soon as they are on their way. Thread the
composer's onPending through the command dispatch so a sent /compact
or /clear leaves the input immediately; a refusal still hands it back.
Fixes#39122
* fix(native-chat): keep a command typed while a refused one was answering
A refused conversation command is handed back through the independent
draft merge, which replaces a lone command in the composer. With the
draft now released on send, a /clear typed while /compact was answering
was overwritten by the refused /compact. Return a refused command only
to an empty composer; anything typed since stays as typed, and the
refusal notice still says the command was not done.
* fix(native-chat): count images when a refused command checks for an empty composer
---------
Co-authored-by: mahedi-emon <113725370+mahedi-emon@users.noreply.github.com>
* feat(dictation): remove filler words from dictated text, per language
Spoken fillers (um, uh, äh, えーと, 음, 嗯 and their equivalents) are
dropped from dictated text before it is inserted, on desktop, the web
client and phones that dictate through the computer. The pass runs once
in the speech host's event sink, so every route shares it, and lives in
a new shared text pipeline that later passes join.
Only standalone sounds go, never words with meaning, and the comma after
a filler goes with it. The language comes from a single-language model
or the dictation language setting; an unknown language uses only
fillers that are no word in another language. Voice settings gain a
Remove Filler Words switch, on by default.
* fix(dictation): take the filler language from what the engine is told, and drop the comma before a final filler
* fix(dictation): keep 嗯 as a reply, clean all-caps transcripts, and read the sentence start in constant time
* refactor(dictation): one list of sentence-end marks in filler removal, and drop a duplicate off-switch test
* fix(dictation): write non-ASCII filler punctuation as escapes so the orcad bundle stays ASCII