The terminal ran on the ghostty-web npm package plus a hand-written patch,
and every rendering bug (recycled rows, duplicated reflow fragments, prompt
artifacts) had to be worked around from outside. The emulator now is the
official libghostty-vt C ABI compiled to WebAssembly, driven by a browser
adapter ported from T3 Code (MIT, notice in LICENSE-T3CODE) and owned in
packages/ui/src/lib/ghostty. The artifact is reproducible with
scripts/build-libghostty-wasm.sh, including a workaround for Zig 0.15.2 on
macOS 27 SDKs.
On top of the port: one WASM instance per page with every tab kept mounted
and hidden tabs paused; history replayed at the PTY size it was drawn for;
shells spawned only after the first fitted grid so zsh never prints the
PROMPT_SP marker; box drawing, block elements and Powerline arrows drawn
procedurally to the exact cell so TUI borders and block logos have no gaps
between rows; a software-rasterized canvas so Gecko renders every tab's text
with the same smoothing; the symbols-only Nerd Font bundled instead of a CDN
fetch; touch selection and scrolling driven through the surface API; a copy
button in the tab strip for touch hosts; localized aria labels.
Testing: bun tests run the real WASM (reflow, palette, replay isolation,
recycled rows, box glyph geometry); viewport and view tests use a surface
double; verified in Chromium and Zen (windowed and headless) for crisp text,
new tabs, panel reopen, resize and box glyph rendering; package type-check,
oxlint/eslint on new files, web build.
Remove automatic moves on session activation, terminal failures, and archive restoration while preserving manual moves and worktree deletion.
Replace directory listing probes with a stat-only endpoint using Node built-ins, including an isolated module-load regression test for packaged desktop.
Validation: focused session, worktree, filesystem, localization, and bridge tests; workspace type-check and lint; web and VS Code builds. Desktop startup and behavior verified by the maintainer.
ghostty-web 0.4.0 hands rows that scroll into view out of recycled WASM
page memory without clearing them, so after a tab or project switch the
new emulator showed the previous terminal's text (upstream #138). The fix
landed only in prereleases, so pin 0.4.0-next.20 and carry the local
block-glyph rendering patch over to the new dist file.
Verified in a production build: creating an emulator after disposing a
full one no longer exposes its rows, and switching between two projects
with live output in each keeps every terminal's content to itself.
Opening the terminal panel sometimes showed stray fragments on the prompt
row: zsh's end-of-line mark and pieces of the prompt path. The shell had
laid its output out for one PTY width, but the client replayed that
history into an emulator of another width (an early size estimate, a
remount, or a renderer rebuild after fonts loaded). ghostty-web's reflow
then left fragments the shell's SIGWINCH redraw never clears.
The server now reports the PTY cols/rows in every snapshot, the transport
carries them through projections and accepted resizes, and the viewport
replays a sized snapshot chunk at that size before returning to the
fitted size. The container-based size estimate only seeds newly spawned
shells and is no longer sent to a running PTY.
Tests cover the sized replay, the store chunk size, the transport
projection, and the server snapshot; verified in a production build by
reloading with the panel open and switching tabs at a changed width.
Add optional completed-turn statistics without changing the existing panel layout. Separate final text delivery speed from whole-turn throughput, preserve scope and opt-in settings, and explain each metric with localized delayed tooltips.
Validated focused telemetry, lifecycle, sync and persistence tests, all-workspace type-check and lint, web builds, the 12-locale narrow layout, and full GitHub CI.
* fix(sessions): keep a shared chat directory until its last session is deleted
Deleting a root chat session removed its managed scratch directory even
when forks, side threads, or subagents still lived in it; OpenCode then
failed every prompt in those sessions with FileSystem.realPath NotFound.
The directory is now removed only once no other known session resolves
to it. The deleted subtree does not count, because the server cascade-
deletes it, and an unloaded global cache keeps the directory instead of
guessing.
Closes#3312.
* fix(sessions): relocate a session whose worktree directory disappeared
A worktree removed outside OpenChamber, by the agent or by hand, left its
sessions pointed at a path that no longer exists: every terminal create
and restart failed with "Invalid working directory" and the tab stayed
stuck, while Git, Files, and prompts kept targeting the dead path.
The terminal server now names that one rejection (TERMINAL_CWD_MISSING)
instead of substituting a directory of its own. The shared UI reuses the
archived-restore fallback for live sessions: a server-confirmed missing
directory moves the session and its stranded subtree to the project's
primary directory through the control-plane move, clears the worktree
hint, re-selects the session, and tells the user where it went. It runs
from a terminal failure and on activation of any session whose directory
is neither a project root nor a managed chat directory; available,
unknown, and failed probes leave everything untouched.
Closes#3338.
* fix(scripts): make oc-dev load again after the changelog cleanup
The changelog cleanup referenced fs.existsSync in a module that imports
existsSync by name and never binds fs, so every oc-dev invocation failed
with "fs is not defined" before reaching its action.
* fix(sessions): probe directory availability on disk, not through OpenCode path resolution
OpenCode's /path never checks that a directory exists: it echoes the
requested path and resolves its project through Git discovery that
swallows errors, so a deleted worktree came back as a valid location and
every missing-directory fallback (draft recovery, archived restore,
session relocation) stayed inert on a real server. The probe now asks
OpenChamber's own /api/fs/list, which stats the path and reports
not-found and not-directory explicitly; anything else stays unknown.
* fix(sidebar): keep a worktree whose directory is gone visible as missing
git keeps a worktree registered after its directory is deleted outside
git and marks it prunable; the list parser ignored that line, so a
deleted worktree looked alive, and nothing in the app asked for a new
listing anyway. The server now reports prunable, the UI keeps such a
worktree in the topology with worktreeStatus missing and a warning icon
on its sidebar group, and relocating a session out of a confirmed-
missing directory raises an in-app topology signal the sidebar
rediscovers on. Dropping the worktree instead would hide every session
that lived there, and a hidden session can never be opened or relocated.
No idle polling is added.
* fix(sessions): never relocate a session to the filesystem root
OpenCode files a directory outside any Git repository under its global
project, whose worktree is the filesystem root. A managed chat whose
directory vanished would otherwise be moved to /. The relocation now
refuses a root destination, and the activation probe recognizes chat
directories through the home-based check as well, so it does not depend
on the chats root having been resolved yet.
* test(sessions): mirror the relocation action in the issue-2039 session-actions mock
session-ui-store now imports relocateSessionFromMissingDirectory, and the
mocked module in this test listed every other action but not that one, so
the file failed on import.
Input history (#3035) shipped with "All projects" as the default scope and
only recorded prompts sent after the upgrade, so ArrowUp showed other
sessions' prompts and, once switched to "Current session", nothing at all.
Default to the current session and merge the visible transcript's prompts
with the persisted bucket. Existing sessions recall as they did before
#3035, while new prompts keep their attachments and stay recallable after
a revert hides them from the transcript.
Cleanup across #1855, #2297, #3072, #3178, #3035 and #3135: drop the
duplicate poll guards in the file content poller, the zod schema the
VS Code package cannot depend on, a copied file-URL helper and stray
whitespace; move the Enter-to-send strings into the settings namespace;
document OPENCHAMBER_CHATS_DIR, resolve the chats root once on the server
and warm it alongside the other bootstrap calls.
* feat(chats): relocate managed chat worktrees via OPENCHAMBER_CHATS_DIR
Projectless-chat worktrees were hard-pinned to
<home>/.config/openchamber/chats: the UI joined the path client-side,
workspace checks allowed only the config root, and identification matched
the literal path segment. When the OpenCode server runs as a separate
user (UID-separated setups), that root is unreachable — every chat
session answered HTTP 500 (EACCES on the session directory).
The server now owns the chats root. OPENCHAMBER_CHATS_DIR relocates it
(default unchanged: <config root>/chats); /api/fs/home answers
{ home, chatsRoot }; fs workspace checks accept the managed chats root
next to the config root; the client resolves the root from the server
(per-runtime cached, warmed at bootstrap so sync classification sees it)
and falls back to the home join for older servers.
Refs #3130
* chore: trim added comments to local precedent
* fix: forward managedChatsRoot through feature-routes-runtime to registerFsRoutes
* fix(chats): await the root warm-up and keep the legacy chats root owned
Review feedback on #3135:
- bootstrapGlobal now awaits warmChatsRootDirectory, so synchronous
session classification never sees an empty root cache (relocated
sessions were grouped as project sessions when the session list
outran /api/fs/home).
- managedProjectRoots keeps the legacy <config root>/chats entry next to
OPENCHAMBER_CHATS_DIR, so memory ownership of existing chats survives
relocation.
* fix(chats): distinguish chats-root fetch failure from older servers
* fix(sync): rehydrate managed chat sessions after the chats root warms
* fix(fs): pass managed roots through the symlink and git-dirs path checks after the main merge
* docs: drop changelog edits; changelog is the maintainer's release-time work
* fix(chats): keep legacy chat directories deletable while the root is relocated
* fix(chats): resolve roots before cleanup and initial session loads
* test(chats): type runtime spies against actual SDK contracts
---------
Signed-off-by: Steffen Mächtel <info@steffen-maechtel.de>
Co-authored-by: Bohdan Triapitsyn <artmore@protonmail.com>
* feat(ui): add composer enter-to-send toggle and native hardware-keyboard detection
Replaces the settings-page "Enter sends with a keyboard attached" checkbox with
an EnterKeyToggle in the composer footer: plain Enter submits / Shift+Enter
inserts a newline when enabled, Shift+Enter submits / Enter inserts a newline
when disabled. Ctrl/Cmd+Enter always submits as the soft-keyboard fallback.
Persisted as enterToSend.
Adds the Android HardwareKeyboardPlugin: scans input devices for an alphabetic
physical keyboard (ignoring phantom key/sensor devices), re-answers on config
changes/foreground, and confirms attachment from real hardware key events.
MainActivity surfaces key events to it before the WebView consumes them. The
composer and draft layout start keyboard-aware instead of inferring one focus
late; ComposerEditor preserves Enter modifiers through CodeMirror's deferred
re-dispatch so the toggle can tell Shift/Ctrl+Enter from plain Enter.
Removes the settings search entry and i18n keys for the old checkbox.
* refactor(ui): keep enter-to-send branch focused
* fix(ui): preserve enter-toggle taps on touch
* fix(ui): preserve enter key defaults and move setting
* fix(ui): keep enter setting lint-clean
* fix(i18n): preserve current Turkish message parity
* fix(settings): persist enter-to-send preference
* fix(ui): clarify enter-to-send setting
* fix(ui): apply enter preference on desktop
* fix(ui): match enter setting focus mode default
* test(ui): cover enter key policy matrix
* fix(ui): harden deferred enter handling
* fix(chat): preserve untouched Enter policy and validate settings
---------
Co-authored-by: Bohdan Triapitsyn <artmore@protonmail.com>
* fix(files): guard file polling races
- Ignore sub-millisecond mtime jitter on a same-size file so an unchanged
open file no longer loops through reload and flickers.
- Swap externally changed text content into the open editor in place
instead of clearing the loaded path and showing the load spinner.
- Read content only after metadata changed, confirm it with a second
read, and skip the swap when the file is unchanged, the buffer is
dirty, or a newer local write landed.
- Keep the stat baseline unchanged when a poll cannot observe content so
a failed read is retried rather than treated as unchanged.
- Fall back to a full reload for images, PDFs, binaries, and files above
the content-poll byte limit, and when a poll returns binary content.
- Serialize polls and dispose the poller on unmount, file switch, and
directory change.
- Add a `fresh` file read option that bypasses the content cache and the
HTTP cache.
* fix(files): invalidate stale polls after diagram saves
---------
Co-authored-by: Bohdan Triapitsyn <artmore@protonmail.com>
Release notes now carry New, Improvements, Fixes, and Misc headings.
Inside the update dialog they render as small uppercase labels over
their lists instead of full-size document headings.
Claude-Session: https://claude.ai/code/session_01VqV56Hez25hTxXH4ipJfzH
The splash and the UI bundle both started in English on a fresh install
even when VS Code ran in a supported language, because only OpenChamber's
own saved locale was consulted. The HTML now exposes VS Code's display
language; the splash and detectInitialLocale use it until the user picks
a locale, which still wins.
Claude-Session: https://claude.ai/code/session_01VqV56Hez25hTxXH4ipJfzH
While a turn runs, the collapsed pill showed a plain Send button for a
draft that the expanded composer shows as Queue with a rotated icon, and
the pill routed it through the primary action. The pill now queues, with
the same label and icon as the expanded composer.
Claude-Session: https://claude.ai/code/session_01VqV56Hez25hTxXH4ipJfzH
The sidebar indicator kept the terminal listing loop going for as long as
the sidebar was visible, in every client, whether or not anything was
running. It now lists once on mount, to pick up runs another client
started, and keeps the loop only while a project action is known to be
running anywhere; an idle sidebar costs no polling.
Claude-Session: https://claude.ai/code/session_01VqV56Hez25hTxXH4ipJfzH
The terminal store keyed its directories by trimming trailing slashes,
while the new sidebar activity indicator and the whole-server session
grouping keyed the same folder with the project-action normalizer, which
also rewrites backslashes. On Windows that split one project into two
namespaces: the sidebar reconciled server sessions under "C:/repo" while
the terminal panel and the project actions button worked under "C:\repo",
so the running-action indicator never lit up, adopted tabs were duplicated
and a stop recorded in one namespace did not guard reconciliation in the
other. The project actions button also read the store map directly with a
path normalized somewhere else, missing its own directory entry.
All terminal directory keys now come from normalizeTerminalDirectory in
lib/pathNormalization, and store reads go through getDirectoryState.
The 256 KiB budget for an oversized tool diff preview counts UTF-16 units, so
the cut can land between the two halves of an astral character - an emoji in a
patched string, for instance. The preview then ended in a lone surrogate that
renders as the replacement glyph.
Step back one unit when the boundary splits a pair.
The IME fix covered the diff/file comment input and the browser annotation
overlay, but two other places where a user writes a comment still acted on
the Enter that confirms an IME candidate: the chat quote comment in the text
selection menu, and the in-place comment editor on a composer context chip.
Confirming a candidate there attached or committed the half-typed reading and
closed the input. Escape, which abandons a candidate, had the same problem.
Both handlers now return early on a composing keystroke, and a single test
asserts the guard across every comment input so the next one added does not
quietly skip it.
Child discovery now pages through the whole session list, so one pull can
take longer than the 15s watchdog interval. Two overlapping pulls each read
the store before either commits, so both appended the same newly found child
and the directory store ended up with duplicate rows. Keep one pull per
directory in flight, mirroring the status-poll guard.
* feat(chat): render code comments as cards instead of fenced text
* fix(vscode): route Add Comment to the active session editor panel
* fix(chat): persist queued inline comments and tighten file-chip path matching
* feat(vscode): comment on code from the editor
* fix(chat): keep attached context in the message and broadcast comment removal
* fix(vscode): hold every pending comment and gate both entry points on the workspace
* fix(vscode): let only the owning surface decide its comment threads
* fix(vscode): drop a comment removed while its delivery was still in flight
* test(vscode): cover the in-flight comment removal guard
* test(vscode): cover comment removal reaching every chat surface
* fix(vscode): give up on a comment the chat never confirmed holding
* fix(vscode): retract a comment everywhere before reporting it discarded
* fix(chat): preserve queued comment cards
* fix: preserve inline comment context across send paths
* fix(chat): preserve command routing with context
* fix(chat): keep unavailable actions on normal send path
---------
Co-authored-by: Bohdan Triapitsyn <artmore@protonmail.com>
When the server delivered the last item of a session's queue it deleted
the session entry and broadcast the update with an empty directory. The
UI keys its projection by directory, could not build the key, and dropped
the event, so the delivered message stayed in the "Queued messages" card
until the next full hydration.
The server now remembers a session's directory beyond the emptying of its
queue so that broadcast names it. The UI additionally treats an empty
session without a directory as "this session's queue is done" for every
projection keyed under that session id, which covers clients talking to
an older server.
Claude-Session: https://claude.ai/code/session_01VqV56Hez25hTxXH4ipJfzH
Thanks for extending the existing IME handling to comment inputs and documenting the browser-event checks. We will also protect the annotation Escape handlers from cancelling composition.
Thanks for fixing the cramped table columns and covering the streaming-to-settled transition. The content-sized layout and horizontal scrolling look ready to merge.
Thanks for refining the action placement. This keeps New chat available while idle and preserves Abort while a response is running. We will finish the repository-specific cleanup in the same batch.
Settings → Integrations offered install cards for the Claude Code and
Cursor provider plugins. They are gone: the section, its plugin catalog,
its own i18n module and tests, the settings search entries, the page
keywords, and the two sprite icons only it used. The page now holds the
built-in GitHub and Linear cards, so it is hidden in VS Code where neither
applies; its title and description live in the settings dictionaries.
Docs follow: the Integrations page in every locale now documents GitHub
(pointing at its own page) and Linear in full, the GitHub page names
Settings → Integrations as the place to connect and covers linking an
issue or PR to a message, and the Providers pages no longer promise Claude
or Cursor subscriptions.
Claude-Session: https://claude.ai/code/session_01HB9wdLQoZX2vfyDjwv6Rso
Queueing captured only the text and files. Context chips (inline comments,
terminal selections, browser annotations, PR comments and checks, quotes,
linked issue/PR/Linear references, pending synthetic parts) stayed in the
composer and only left with the next manual send, so a queued message the
server delivered went out without them and the chips rode an unrelated
message later.
A queued message now carries what the composer would have sent: the text
with its agent mention stripped and file mentions resolved into
attachments, the attached context as structured parts, and the skill
instruction derived from the text. The server delivers those parts in the
composer's order, the VS Code auto-send does the same, and editing a queued
message puts the chips and linked references back. A failed queue restores
the composer completely. Snapshots and broadcasts omit the captured
context like attachment payloads; a take returns it.
Claude-Session: https://claude.ai/code/session_01HB9wdLQoZX2vfyDjwv6Rso
The send write was fixed, but the message that send echoes back arrives with
model metadata and no effort, and its model matches the one the send just
saved. That is exactly when `shouldPreserveManualModelOverride` declines to
protect the selection, so the history branch ran with no variant and recorded
"no choice" over the user's explicit `Default`. Picking Default, sending, then
switching agent and back still put the settings default in the picker.
A message carrying no effort is not evidence that the session has none. The
history branch now keeps whatever the session already recorded when the message
carries no variant, and a concrete historical effort still replaces it.
The behavior test ran the real guard through a mock that returned a fixed
answer, so the failing branch had no coverage. It now calls the real function
unless a test opts out, and the new case fails without this change.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EZuVVgziiLjD81W5vaxdH2
A send carries no effort both when nothing was chosen and when the user picked
"Default", so `sendMessage` could not tell the two apart and recorded the raw
value, which clears the entry. Picking "Default", sending, then switching agent
and back put the settings default back in the picker — the shape of the bug
this branch set out to fix.
`materializeOpenDraftSession` already read the live selection to keep that
distinction on the draft path. Both paths now share `resolveVariantToRecord`,
which prefers `currentVariantSelection.override` while the live selection still
describes the agent and model being sent to, and falls back to the sent value
when it does not.
Three tests go through the real `sendMessage`; two of them fail without this
change. The existing ones seeded the record directly, which is why the send
path was never covered.
Also documents the `oc.chatInput.lastDraftTarget` record in the owning sync
documentation: its three `target` values, what a pre-`target` record and a
removed project fall back to, and why a chat scratch directory is not a project
target.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EZuVVgziiLjD81W5vaxdH2
Nothing covered ModelControls, which is where the effort restore decides
between "the user chose Default" and "nothing was found". Three cases: a
concrete effort in the session history is restored, history without an effort
records no choice, and a preserved manual override keeps a recorded Default.
The middle one fails on the previous commit's parent — it recorded `null` and
then `undefined` for the same restore.
Also drops the `fetchAll` binding GitView stopped using when the post-bootstrap
refresh moved to a forced `fetchStatus`; it was failing `bun run lint:ui`.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EZuVVgziiLjD81W5vaxdH2
`commitVariantSelectionForModel` turned every `undefined` into an explicit
`Default` (`null`), but it serves two kinds of caller. The picker means "the
user chose Default"; the history and manual-override restores mean "nothing
was found". Restoring a session therefore recorded a choice nobody made, and
`resolveModelVariantSelection` collapsed that `null` back to `undefined`, so
the next restore recorded it again. Because an explicit `Default` outranks the
agent and settings defaults by design, the session latched onto `Default` and
the concrete effort its own history carried could not come back.
Move the decision to the callers: the four picker paths pass `variant ?? null`,
the restore paths pass their result through, and the resolver returns the
selection store's three states instead of two. The follow-up write in the
history restore goes with it — the apply above it already recorded the same
agent and model, and a second write could only disagree.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EZuVVgziiLjD81W5vaxdH2
Bohdan Triapitsyn