03697190ffe886bdbe1352b23d352c06ef0b2e7d
11
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Bohdan Triapitsyn
|
687681c83b |
fix(providers): complete OAuth logins that finish in the browser
OpenCode's authorize response reports how the client must finish: `code` expects a pasted code, while `auto` requires the client to call oauth/callback immediately and hold it open — upstream blocks in there polling for the device code or waiting on its loopback redirect, and only that call persists the credential. Every auth plugin OpenCode ships uses `auto`; none use `code`. The page implemented only `code`. It opened the browser, showed a paste field no provider can fill, and never called back, so a successful sign-in stored nothing and the app sat unchanged. Authorization now drives the UI: `auto` chains straight into the callback behind a waiting state with a cancel, and the paste field appears only when a provider actually asks for a code. Two smaller failures shared that surface. Prompts were never collected, which put GitHub Copilot Enterprise out of reach entirely, so a method that declares them now asks first and passes the answers to authorize. Device codes are also recovered from the instructions text, where they actually live — the old code read fields the API does not return, so the copy button never appeared. The callback is exempt from the ordinary proxy deadline and gets a 15-minute budget, bounded by the shortest upstream expiry we know of. A human sign-in with 2FA does not fit in four minutes, and expiring it turned a completed login into a 504. |
||
|
Bohdan Triapitsyn
|
0f830f8804 |
fix: stream bash output and harden OpenCode connectivity (#2522)
* fix(ui): stream bash tool output while running * perf(ui): render streaming bash output incrementally * fix(ui): keep tool duration timer running * fix(web): recover stalled OpenCode SSE streams * fix(web): prevent OpenCode restart storms * fix: address streaming recovery review |
||
|
Bohdan Triapitsyn
|
40dfff4a9a | fix: handle ambiguous prompt transport failures | ||
|
Bohdan Triapitsyn
|
e372c8d8cb |
perf: instant startup via cache hydration + decoupled readiness (#1650)
* perf(startup): hydrate providers/agents from cache (stale-while-revalidate)
Persist last-known provider/agent snapshots instead of stripping them, so the
model/agent pickers paint instantly on cold start. Freshness is preserved by the
background refresh in initializeApp() and activateDirectory() (which overwrite on
success) and by the existing provider/agent config-change subscriptions, so the
prior stale-provider regression stays fixed without blanking the UI during fetch.
* perf(startup): cache directory session list for instant sidebar
Persist a capped slice of each directory's session list and seed the child store
from it on creation, so the sidebar paints chats immediately on cold start.
Bootstrap phase-3 loadSessions overwrites with the fresh list; its empty-list
race guard preserves the seeded sessions during OpenCode warmup.
* perf(startup): hold API requests through OpenCode warmup instead of 503
The readiness gate returned 503 the instant OpenCode wasn't ready, pushing the
client into an exponential-backoff retry loop (500ms -> 1s -> ...) that wasted
seconds of cold-start time and could fail bootstrap outright. Now hold the
request and poll readiness up to a bounded window so the first call succeeds as
soon as OpenCode is up (typically sub-second); still 503 fast past the window so
a genuinely-down server doesn't hang. Adds coverage for both paths.
* perf(startup): surface cached providers/agents in pickers (optimistic readiness)
The model/agent pickers gated purely on isInitialized, so they showed
"Loading…" for the entire init round-trip even when provider/agent data was
already hydrated from cache — making the persisted-cache work invisible. Treat
the pickers as ready as soon as cached providers are present (stale-while-
revalidate), so they paint last-known models/agents instantly and refresh in the
background. First-ever launch (no cache) still shows Loading until init.
* perf(startup): don't abort directory bootstrap on transient phase-1 failure
A failed initial path.get OR session.status aborted the whole directory
bootstrap, stranding it in loading and skipping phase 2/3 (session load).
session.status is live data the event pipeline keeps current, and path.get is
tolerable once a project is resolved from global state. Now only a total
failure (or path.get failing with no resolved project) aborts, so the sidebar
and chat keep advancing and loading sessions through warmup hiccups.
* perf(startup): don't bootstrap directories from archived sidebar rows
Each sidebar session row called useDirectoryStore(dir), which defaulted to
bootstrap:true and triggered a full directory bootstrap. Archived sessions point
at dozens of (often deleted) worktrees, so on startup this fired a session-list
fetch + 6x2s empty-retry storm per dead directory (the logs the user saw). The
store ref there is only read on-demand via getState() in export handlers, never
subscribed, so archived rows don't need it bootstrapped. Add a { bootstrap }
option to useDirectoryStore and skip bootstrap for archived rows; active rows
still bootstrap so live cross-directory session/status keeps aggregating.
* perf(startup): stop empty-session bootstrap retry storm on web/desktop
The post-bootstrap retry re-ran the full directory bootstrap 6x2s whenever the
session list came back empty, on the theory that empty meant OpenCode wasn't
ready. But loadSessions already retries transient failures twice over
(listGlobalSessionPages throws on 5xx and retries internally), so on web/desktop
an empty result is authoritative — the directory genuinely has no sessions (e.g.
deleted worktrees referenced only by archived sessions). That produced the
dozens of '[bootstrap] sessions empty ... 6 attempts; giving up' log storms.
Gate the retry to VS Code, where the bridge can return an empty 200 during
warmup that the inner retries can't catch.
* perf(startup): scope provider/agent config to project (worktrees inherit)
Providers/agents/defaults are project-level, but were keyed per directory, so a
worktree fetched and cached its own snapshot — duplicating the parent project's
load (the trace showed initializeApp loading the worktree and activateDirectory
loading the project concurrently, ~8s of redundant background work).
- resolveConfigDirectory() maps a worktree to its owning project; loadProviders
/loadAgents/activateDirectory now key by it, so a worktree reuses one shared
project snapshot. activateDirectory resolves up-front so activeDirectoryKey and
the snapshot key always match (picker stays consistent); the OpenCode working
directory is unaffected.
- Add a 30s runtime freshness guard so the stale-while-revalidate background
refresh skips re-fetching config that was just loaded (initializeApp then
activateDirectory for the same project), and to avoid churn on rapid project
switches. Config-change invalidation clears the snapshot, which bypasses the
guard, so freshness never masks a needed refresh.
* fix(sidebar): default archived sessions to hidden to avoid startup flash
useSessionDisplayStore defaulted showArchivedSessions to true, so on startup
archived sessions rendered by default and then vanished once the persisted
preference rehydrated to hidden — a visible flash. Default to hidden so the
pre-hydration state is the quiet one; users who opted into showing archived keep
their persisted true (default change doesn't override persisted state).
* perf(startup): persist worktree->project mapping to kill cold double-load
The worktree->project map (availableWorktreesByProject) is populated by async git
discovery, so it isn't ready when initializeApp runs — a worktree's first config
load couldn't resolve to its project and duplicated the project's provider/agent
load, saturating OpenCode during cold start (the source of the slow first
createSession/send the user observed). Cache resolved worktree->project mappings
to localStorage so resolveConfigDirectory resolves synchronously at init on
subsequent launches; the project is loaded once and activateDirectory hits the
freshness guard. worktree->project is immutable so a cached entry is safe; live
resolution still populates/corrects the cache.
* perf(startup): persist worktree map for instant sidebar + first-launch keying
Worktree discovery is async (git), so availableWorktreesByProject was empty at
startup: the sidebar worktree list appeared late, and useConfigStore couldn't
resolve a worktree to its project on the first launch (causing the cold
worktree+project double-load). Persist the discovered worktree map to
localStorage and seed it synchronously on store init (stale-while-revalidate:
discovery refreshes in the background via the existing setState, which now
write-through persists). The sidebar paints worktrees instantly and
resolveConfigDirectory resolves the project from the very first launch.
* perf(startup): coalesce concurrent duplicate OpenCode reads in runtimeFetch
On cold start the sync bootstrap and the config store independently fire the same
idempotent reads (providers, config, path, agents, project) concurrently with no
shared dedup, saturating the single OpenCode process and delaying work queued
behind it (e.g. createSession). Coalesce genuinely-concurrent identical GETs to
those read endpoints at the transport layer so OpenCode does the work once; each
caller receives an independent response clone. Tightly scoped: GET only,
allowlisted read paths, never event streams, never a signal-bearing request (so
one caller's abort can't cancel the shared fetch). Entries clear on settle, so it
only shares overlapping in-flight requests — never a stale response.
* perf(startup): cache git branches so the draft branch selector paints instantly
The branch selector above the composer was the slowest-loading element: it's
gated behind a cold 'git branch' fetch (useGitStore, not persisted). Cache the
per-directory branch list to localStorage and seed the store on init (with
isGitRepo:true so the selector's gate passes), and write the cache on every
successful fetchBranches. The ChatInput draft-branch effect now refreshes on
staleness (>30s) rather than mere absence, so seeded branches show immediately
and still refresh in the background without a spinner — no stale-forever
regression. Only the branch list is cached; status/log/diff are untouched.
|
||
|
Hristo KaramanlievandBohdan Triapitsyn
|
9685630436 |
fix: lighten session list payloads (#1538)
* lighten session list * fetch full session on open * sanitize session list * sanitize global sessions * preserve revert markers in session lists * fix: preserve session metadata in list sanitizers --------- Co-authored-by: Bohdan Triapitsyn <artmore@protonmail.com> |
||
|
Bohdan Triapitsyn
|
2031e3b4a8 |
Decouple bundled UI from runtime API and add remote instance tooling (#1228)
Add a packaged-client runtime boundary so the shared UI can talk to local, desktop, remote, and VS Code runtimes through the right transport instead of assuming one same-origin web server. Centralize OpenChamber-owned API access behind RuntimeAPIs, runtimeFetch, and runtime URL helpers, while keeping official OpenCode traffic on the SDK path. Support runtime switching, remote host selection, desktop client credentials, and headless connection links for pairing packaged clients with remote OpenChamber servers. Harden the new auth model by moving long-lived client tokens out of browser URLs, introducing short-lived scoped URL tokens for browser-owned transports, restricting URL-token access to explicit readable/realtime routes, and making client-token management session-scoped or self-scoped as appropriate. Update browser-owned assets and preview proxy flows to work with the split runtime model, including authenticated project icons, preview token propagation, CSP-safe preview bridge injection, and preview proxy auth that survives short-lived URL-token expiry. Tighten Electron security boundaries for packaged clients by gating privileged preload state to trusted origins and requiring explicit confirmation before connect deep-links import or switch remote runtimes. Also refresh agent guidance and project skills so future runtime/API, auth, preview, UI, CLI, settings, locale, and drag-to-reorder work follows the new architecture. |
||
|
Jinwoo An (안진우)andBohdan Triapitsyn
|
6470d9d205 |
fix: preserve lastEventId in SSE path and add proxy heartbeat (#1041)
* fix: preserve lastEventId in SSE path and add proxy heartbeat - Extract event.id from SSE stream events in event-pipeline.ts so that reconnects carry the correct Last-Event-ID header for gapless replay. - Emit :heartbeat comment every 20s in the direct SSE proxy to keep the UI heartbeat watchdog from aborting idle connections. * fix: handle SSE metadata through SDK callback * Guard SSE proxy heartbeats --------- Co-authored-by: Bohdan Triapitsyn <artmore@protonmail.com> |
||
|
Bohdan Triapitsyn
|
9d9af7e263 | Improve event stream resilience | ||
|
Bohdan Triapitsyn
|
285c3bcaae |
Migrate desktop shell from Tauri to Electron (#964)
* feat(electron): scaffold Electron desktop package
Main + preload + ssh manager, packaging scripts, icons, root build/lint/type-check wiring.
* feat(ui): add Electron runtime detection and desktopNative facade
isElectronShell via window.__OPENCHAMBER_ELECTRON__, isDesktopShell now covers both. desktopNative wraps window/title/theme calls so UI avoids direct Tauri imports. revealDesktopPath added.
* refactor(ui): route window/title/theme/export through desktopNative
SessionSidebar, MultiRunLauncher, useWindowTitle, ThemeSystemContext, exportSession drop direct @tauri-apps imports.
* refactor(ui): treat all desktop shells uniformly
device.ts switches Tauri-only checks to isDesktopShell. Header OpenInApp button uses actionDirectory so it falls back to the active project path.
* fix(ui): menu Copy clipboard fallback and softer sidebar tint
useMenuActions falls back to Clipboard API for the native Copy action when the page doesn't intercept. cssGenerator lowers sidebar strong/soft alpha so the tinted surface reads gentler.
* chore(electron): mirror Tauri build/type-check script shape
build script becomes no-op so root 'bun run build' skips packaging. Syntax validation (node --check) moves into type-check. electron:build root script still runs full sidecar+bundle+electron-builder.
* fix(electron): sync app identity, preload path, boot outcome, dev entry
Read version from packages/electron/package.json so 'electron ./main.mjs' dev entry reports the app version instead of Electron's. Bump electron package to 1.9.6 for workspace parity.
Resolve preload via app.getAppPath() in prod (bundle lives in dist-bundle while preload.mjs ships at app root).
Compute and inject __OPENCHAMBER_DESKTOP_BOOT_OUTCOME__ in main + preload so the loading gate dismisses (mirrors Tauri Rust injection).
Dev entry uses ./main.mjs to bypass the stale dist-bundle so source edits apply.
* refactor(open-in-app): split directory and file flows
Header button now opens the project/worktree directory only — drop activeFilePath prop and its Header prop passthrough. FilesView editor dropdown opens the active file only via new openDesktopFileInApp.
Electron main.mjs mirrors Tauri's open-chain logic: buildOpenProjectSpecs (finder/terminal direct, vscode-like via CLI -n, JetBrains via open -na --args) and buildOpenFileSpecs (finder -R reveal, terminal opens parent dir, editors via CLI or open -a). runSpecChain falls through specs until one exits 0.
* fix(files-view): keep floating toolbar mounted while its dropdowns are open
Portalled Base UI menu popups render outside floatingToolbarRef. The document mousedown listener and onMouseLeave collapsed the toolbar as soon as the popup appeared, unmounting the DropdownMenu root and swallowing clicks on its items. Track open dropdowns via onOpenChange and skip the collapse while count > 0; also ignore mousedowns that land inside a dropdown-menu-content/item.
* feat(electron): add quit confirmation with risk poller
Mirrors Tauri's macOS-only behavior: poll /api/openchamber/scheduled-tasks/status and /api/openchamber/tunnel/status every 5s. If active tunnel or running/enabled scheduled tasks are detected, Cmd+Q / dock Quit / menu Quit shows a native warning dialog listing reasons; otherwise quit proceeds silently.
performConfirmedQuit persists window state, kills sidecar, shuts down SSH, and fires a 1500ms unref'd safety timeout that calls app.exit(0) if the normal quit sequence stalls.
* feat(notifications): fix payload parsing, restore-on-click, session deep-link
Normalize input so both sidecar stdout path (flat) and UI IPC path ({ payload: {...} }) work; previous destructuring missed requireHidden (camelCase) and the payload wrapper so notifications showed with empty body.
Click handler restores the window if minimized, shows it if hidden, and focuses. When the notification payload carries sessionId, emit openchamber:open-session which the App listener routes to setCurrentSession — matches the PWA service-worker deep-link behavior. macOS notifications now also use sound 'Glass' for parity with Tauri.
* chore(electron): bump to Electron 41 + latest updater/context-menu
electron ^38.2.0 -> ^41.2.1
electron-updater ^6.6.2 -> ^6.8.3
electron-context-menu ^4.0.4 -> ^4.1.2
Dev boot verified: main process starts, preload exposes globals, API server + quit risk poller + autoUpdater all initialize without errors.
* fix: keep todo row alignment stable when expanding text
Keep checkbox and action buttons vertically centered in collapsed todo rows
Prevent first todo line from shifting when expanding to multiple lines
* fix: make commit highlights visible and input behavior reliable
Switch commit message field to native textarea for predictable auto-resize
Fix AI highlights append flow so inserted text is applied consistently
Make chat scroll-to-bottom control fully circular
* style: increase chat bubble corner radius consistency
Use larger radius for user chat message bubbles
Match chat input container radius to user message styling
* feat(electron): adopt OpenCode playbook improvements
mac: hardenedRuntime + entitlements.mac.plist + notarize + dmg.sign for Apple notarization parity.
single-instance lock + openchamber:// protocol with session/project/host routing (host switch done fully in main via activateMainWindow).
setAppUserModelId for Win toast identity; proxy-bypass-list switch; chdir(homedir) for Finder-launch cwd safety.
shell env probe (\$SHELL -il -> -l) merged into sidecar spawn; PATH deduped.
electron-log with 5MB rotation + 7-day cleanup; autoUpdater.logger wired; startup info log.
webContents zoom locked to 1 (zoom-changed + did-finish-load).
UI: openchamber:open-project -> useDirectoryStore.setDirectory.
* fix(electron): make bootOutcome mutable across re-navigation + project deep-link
host deep-link used to land on chooser because contextBridge exposed bootOutcome as read-only; initScript re-assignment became a silent no-op. drop preload's contextBridge for bootOutcome, inject it via main-world initScript, and move injection from did-finish-load to dom-ready so it lands before React mounts.
project deep-link updated currentDirectory only; activeProjectId stayed stale so the sidebar didn't highlight the new project. switch to projectsStore.setActiveProject (or addProject for new paths) which updates both.
add log.info around deep-link dispatch + host switch for diagnostics.
* fix(electron): desktop_hosts_set IPC args + persist initialHostChoiceCompleted + re-eval bootOutcome
UI calls invoke('desktop_hosts_set', { input: {...} }) but main was reading args.config — every onboarding 'i've completed installation' / host-dialog save wrote nothing, so desktopDefaultHostId stayed null and the chooser screen looped forever.
also:
- writeDesktopHostsConfig now persists desktopInitialHostChoiceCompleted so the tauri-compat flag survives writes.
- readDesktopHostsConfig returns initialHostChoiceCompleted so the UI-side config mirror is complete.
- after writing hosts, recompute state.bootOutcome + state.initScript; a subsequent window.location.reload() picks up target=local/status=ok via dom-ready injection without needing a full app restart.
- app.setName('OpenChamber') early (pre log.initialize) so electron-log logs land in ~/Library/Logs/OpenChamber/ instead of the package-derived '@openchamber/electron' path.
* chore(electron): rename appId to dev.openchamber.desktop
ai.opencode.* is the OpenCode team's reverse-DNS namespace; OpenChamber should not squat there. now that we're on Electron, drop the tauri-era inherited identifier and claim our own under openchamber.dev.
user-facing productName stays "OpenChamber". tauri identifier left as-is — legacy shell on the way out.
* feat(ci): add electron build+notarize+publish jobs to release workflow
three new jobs in release.yml, running in parallel with tauri:
- build-desktop-electron-macos: matrix(arm64, x86_64) on macos-26; installs Developer ID via keychain, runs build:sidecar + bundle:main + electron-builder --mac --arch <> --publish=never (with APPLE_ID / APPLE_APP_SPECIFIC_PASSWORD / APPLE_TEAM_ID env mapped from existing secrets). verifies hardened runtime, stapled notary ticket, required entitlements. uploads DMG/ZIP/blockmaps to the release and emits per-arch latest-mac.yml as a GH artifact.
- combine-electron-manifests: downloads latest-yml-*-apple-darwin artifacts, runs the existing finalize-latest-yml.mjs to merge per-arch files entries into a single latest-mac.yml, uploads combined yml to the release.
- finalize-release: now also waits on the two new jobs before flipping the draft release to published.
also: explicit artifactName in electron-builder config so arm64 and x64 dmg/zip never collide.
electron-updater in main.mjs (setFeedURL btriapitsyn/openchamber) fetches this latest-mac.yml on desktop_check_for_updates; downloadUpdate / quitAndInstall wire through our existing IPC handlers unchanged.
* docs: future-agent brief for tauri -> electron auto-update cutover
self-contained plan for the one-shot migration release that carries existing tauri installs into the electron shell via tauri's updater. written so a fresh agent with no branch context can execute it.
covers: the trick (repackage signed electron .app as a tauri tarball, minisign with existing TAURI_SIGNING_PRIVATE_KEY), workflow surgery on release.yml, rollback plan, validation steps against a real tauri install, and edge cases (CFBundleIdentifier change, notification perms re-prompt, deep-link re-registration).
* docs: soften framing of cutover playbook (no user-shaming)
* chore: mark electron as primary desktop shell; tune dmg installer window
AGENTS.md: explicit note that new desktop work lands in packages/electron/, packages/desktop/ (tauri) is maintenance-only until the cutover described in docs/TAURI_TO_ELECTRON_CUTOVER.md. updated runtime/entry-points/build-commands sections accordingly.
electron/package.json build.dmg: cleaner title ("OpenChamber 1.9.6" without -arch suffix), 660x400 window matching the tauri layout users are used to, icon size 128, explicit app/Applications positions.
* refactor(web): drop bun-specific runtime deps from server
- 11 test files migrated bun:test -> vitest; API (describe/it/expect) is drop-in; all 73 tests pass under vitest run.
- bun:sqlite -> better-sqlite3 in git/service.js::syncSandboxesToOpenCodeDb. api shift is db.query().get()/run() -> db.prepare().get()/run().
- add "test": "vitest run" script in packages/web.
no production code used Bun.* APIs; server is Express-on-Node already. this commit removes the remaining bun-runtime shape so the server module can be imported and booted inside an electron main process.
* feat(electron): boot web server in-process, drop sidecar subprocess
the electron main process now imports @openchamber/web/server/index.js as a workspace dependency and calls startWebUiServer({...}) directly. the returned handle exposes getPort() / stop() and the notification emitter takes an onDesktopNotification callback, so we no longer spawn a bun-compiled sidecar binary and no longer parse stdout for the one-line notify protocol.
- packages/electron/package.json: +@openchamber/web (workspace:*); extraResources drops 'sidecar'; build:sidecar script renamed to build:web-assets (kept the vite build step, dropped the bun compile step).
- packages/electron/main.mjs: remove spawn/kill-stale-sidecar/sidecar path resolver/stdout-prefix parser; rewrite spawnLocalServer to probe a free port (stored | DEFAULT_DESKTOP_PORT | OS-assigned) then import server and await startWebUiServer; killSidecar calls handle.stop({ exitProcess: false }); hoist user shell env (PATH, etc.) onto process.env once so opencode / git / rg children still inherit the expected runtime environment.
- packages/web/server/lib/notifications/emitter-runtime.js: accept an onDesktopNotification callback (late-bindable via setOnDesktopNotification). when set, notifications are dispatched through the callback instead of process.stdout; tauri path still uses stdout when no callback is bound.
- packages/web/server/index.js: main() wires options.onDesktopNotification to notificationEmitterRuntime.setOnDesktopNotification.
- release.yml + AGENTS.md updated for the new script name + runtime shape.
payoff: -300ms cold start on mac, single process in activity monitor, no stdio IPC, no bun binary in the packaged app. tauri sidecar path is untouched.
* build(electron): rebuild native deps explicitly, bump electron-builder
the previous build failed because electron-builder 24.13.3 tried to run \`bun rebuild\` on native deps (better-sqlite3, node-pty) and bun has no rebuild subcommand; it also couldn't find prebuild-install because bun hoists under node_modules/.bun/<pkg>@<ver>/ and never populates node_modules/.bin for transitive deps.
fix:
- bump electron-builder devDep to ^26, whose packageManager detection understands bun workspace layouts.
- add @electron/rebuild devDep + scripts/rebuild-native.mjs. the script rebuilds better-sqlite3 / node-pty / bun-pty against the installed electron version before electron-builder is invoked.
- set build.npmRebuild=false so electron-builder no longer attempts its own broken PM-based rebuild.
- package script: build:web-assets -> bundle:main -> rebuild:native -> electron-builder.
verified: CSC_IDENTITY_AUTO_DISCOVERY=false bun run electron:build produces signed-ad-hoc dmg/zip/blockmap/latest-mac.yml; artifacts land under packages/electron/dist as expected. cold-start from Applications should work (native bindings now match electron 41 node ABI).
* fix(electron): externalize web server + native deps from main bundle
the ESM bundle was statically inlining @openchamber/web transitively, which pulled in bun-pty/src/terminal.ts with its top-level \`import { dlopen } from "bun:ffi"\`. node's ESM loader parses every static import when the bundle loads, so the bun:ffi scheme crashed the packaged app at startup with ERR_UNSUPPORTED_ESM_URL_SCHEME — the runtime guard (if (globalThis.Bun) { await import('bun-pty') }) never got a chance to skip it.
fix: bundle-main.mjs marks @openchamber/web (+ its bun-pty / node-pty / better-sqlite3 transitives) as external. the dynamic \`await import('@openchamber/web/server/index.js')\` in main.mjs stays a runtime resolution; the conditional bun-pty import stays dynamic; native modules load from node_modules via the standard resolver.
* perf(web): classify UI-only deps as devDependencies, shrink packaged app
packages/web is a hybrid package: server code in server/, react UI source in src/, compiled UI output in dist/. the server serves dist/ as static files — it never imports react/radix/codemirror/etc. at runtime. but electron-builder, npm install, and similar tools treat everything under "dependencies" as shipping surface, so all of react + @radix-ui/* + @codemirror/* + @fontsource/* + @simplewebauthn/browser + cmdk + ghostty-web + ... were landing in app.asar even though the same code is already baked into dist/ chunks.
move ~24 UI-only packages to devDependencies. vite + its plugins still install them in dev (bun install fetches devDependencies in workspaces), so \`bun run build\` is unchanged. consumers doing \`npm install @openchamber/web\` no longer pull ~150MB of unused browser-side modules.
measured on aarch64 darwin build:
- app.asar: 281MB -> 44MB (-237MB, -84%)
- .dmg: 320MB -> 132MB (-59%)
- .zip: 305MB -> 129MB (-58%)
verified type-check, ui build, 73 vitest tests, packaged launch.
* chore(electron): center dmg installer icons, use cream brand background
dmg-builder 26 ignored our previous dmg.contents positions against its template background (they stayed at template coords, producing misalignment with the drawn arrow). switch to a solid backgroundColor (#FFFCF0, the splash light tone) so the template image is dropped entirely and our coordinates are authoritative. window tuned to 540x340, iconSize 100, iconTextSize 13.
dmgbuild treats contents coordinates as icon *centers* (not top-left), so with iconSize=100 in a 540 window, x=180 and x=360 place left and right clusters with equal 130px gaps on both sides of the window. y=140 vertically centres the icon+label pair.
* fix(electron): eliminate main-thread freezes in in-process server
Three blocking paths were running sync work on the Electron main event
loop, causing multi-second UI freezes under the new in-process server:
- package-manager.detectPackageManagerDetails fired spawnSync(pnpm/npm/
yarn/bun bin -g) with 10s timeouts. In desktop runtime PM detection is
pointless (app is .app bundle, updates via electron-updater) — short-
circuit when OPENCHAMBER_RUNTIME=desktop. This was the ~5s freeze.
- buildInstalledApps iterated 22 OPEN_IN_APPS × spawnSync(mdfind, sips).
Converted to execFile promises so child waits yield to the loop.
- orphan-project-file recovery re-scanned disk on every settings read
(3+/s from fs/list/etc). Cache the outcome per process lifetime.
Also: resolveProjectDirectory prefers settings.lastDirectory over
activeProjectId so file-open from sidebar/chat doesn't 400 with
"Path is outside of active workspace" after the user navigates.
Plus dropdown typeahead fixes in DesktopHostSwitcher/BranchSelector:
stopPropagation on input keys so cmdk doesn't swallow typing.
* feat(electron): restore desktop LAN access for in-process server
spawnLocalServer now reads settings.desktopLanAccessEnabled and binds
on 0.0.0.0 when enabled, so phones/tablets on the same Wi-Fi can open
the app via http://<lan-ip>:<port>. Adds desktop_get_lan_address IPC
(UDP-connect route lookup with networkInterfaces fallback) for the
settings UI to show the reachable URL.
UI and settings plumbing already existed from the sidecar build; only
the Electron main-process wiring was missing.
* chore: added electron package to version bump script
* fix(electron): address PR review — harden IPC surface + polish
P1 security:
- Gate openchamber:invoke and openchamber:dialog:open by webContents
origin. Only local (loopback / dev file://) senders can call desktop_*.
Blocks remote hosts loaded via DesktopHostSwitcher from reading local
files, opening apps, relaunching, etc.
- desktop_read_file now refuses paths outside $HOME / tmpdir and denies
.ssh/.aws/.gnupg/.config/gh/credentials + .env/.pem/.key by name
(defense-in-depth behind the origin gate).
P2:
- webPreferences.sandbox:false: add comment explaining preload needs Node
(contextBridge+ipcRenderer) and why flipping to true would break IPC.
- desktop_set_vibrancy: comment the intentional no-op (no Electron
equivalent for the Tauri NSVisualEffectView path), drop requiresRestart.
- desktopNative.ts: replace isTauriShell() guards with isDesktopShell()
so the semantics match (previous check worked only because Electron
preload exposes a __TAURI__ shim).
- AGENTS.md: correct entry description — server runs in-process, not as
a sidecar subprocess.
* fix(electron): stop leaking desktop shell APIs to remote renderer pages
Preload was exposing __TAURI__ and __OPENCHAMBER_ELECTRON__ unconditionally,
so after DesktopHostSwitcher navigated the window to a remote OpenChamber
instance the remote UI saw isDesktopShell() === true and tried to invoke
desktop_* IPC. The main-process origin gate then threw "IPC not available
for this origin", surfacing as a user-visible error on the onboarding
screen of the remote.
Preload re-runs on cross-origin navigation; compute current origin up
front and only expose the shell globals + the openchamber:emit listener
when the document is loopback / state.localOrigin / file://. Remote
pages now look like a plain web runtime — no IPC path to reject.
* fix(electron): restore remote UI shell integration via per-command gate
Previous commit stripped __TAURI__ / __OPENCHAMBER_ELECTRON__ from remote
pages wholesale, which broke DesktopHostSwitcher for anyone switched to
a remote instance: no hosts list, "Unknown" probe status, open-in-new-
window dead. Also lost window chrome affordances that the remote UI
needs to render correctly inside the Electron shell.
Switch from an origin-level gate to a per-command allowlist:
- preload.mjs exposes __TAURI__ and __OPENCHAMBER_ELECTRON__ on every
page (shell identity + IPC channel). __OPENCHAMBER_LOCAL_ORIGIN__ and
__OPENCHAMBER_MACOS_MAJOR__ also go everywhere since HostSwitcher and
window chrome depend on them and neither grants capability.
__OPENCHAMBER_HOME__ stays local-only (leaks the OS username and is
misleading if consumed as a workspace hint on a remote page).
- main.mjs ipcMain.handle accepts a curated COMMANDS_SAFE_FOR_REMOTE set
(hosts_get, host_probe, new_window, new_window_at_url, set_window_*,
is_window_fullscreen, start_window_drag, get_app_version,
get_lan_address). Filesystem, shell.openPath, installed-apps scans,
app relaunch, auto-update, hosts_set, dialog:open, read_file stay
local-only — remote UI doesn't need them and can't weaponize them.
* ci(release): rebuild native modules against Electron ABI before packaging
Electron job skipped rebuild:native so bun install's Node-ABI builds of
better-sqlite3/node-pty/bun-pty shipped into the asar — packaged app
would crash on require. Local bun run package runs the step via
scripts/rebuild-native.mjs (npmRebuild is disabled in package.json);
mirror it in CI and pass ELECTRON_BUILDER_ARCH so the x64 matrix
cross-builds from the arm64 runner.
Tauri job untouched — both builds continue to produce side-by-side
release artifacts (latest.json for Tauri, latest-mac.yml for Electron)
so each shell's updater finds its own manifest.
* ci(release): split Electron arm64/x64 onto native macOS runners
Both Electron matrix entries were running on macos-26 (arm64) and
cross-building x64 from there. Works for Rust/Tauri; brittle for
native Node modules — better-sqlite3, node-pty, bun-pty (with its
rust-pty crate) each have their own cross-target quirks.
Pin arm64 → macos-14 and x64 → macos-13 so node-gyp and
@electron/rebuild build against the host arch. ELECTRON_BUILDER_ARCH
now just mirrors the runner for clarity.
* Revert "ci(release): split Electron arm64/x64 onto native macOS runners"
This reverts commit f217880e49609cf1418818af0f837b333dbb6f42.
* ci(test-build): add Electron DMG job to arm64 dispatch workflow
Parallel job to the existing Tauri DMG builder, same runner + Apple
cert path. Mirrors the release workflow steps (build:web-assets,
bundle:main, rebuild:native, electron-builder) so maintainers can
smoke-test a signed+notarized Electron DMG before merging.
* ci: use electron-builder v26 boolean arch flags
v26 dropped --arch <name> in favour of per-arch booleans (--arm64,
--x64, etc.). Test build was failing at dispatch time; release job
had the same bug latent. Switch both to the supported form.
* fix(electron): route external links to the system browser
<a href> clicks and window.open calls with non-local URLs were loading
inside the Electron BrowserWindow (or spawning a second Electron window
as a makeshift browser). Add an origin-aware navigation guard to each
window: loopback / state.localOrigin / configured desktop hosts keep
their existing in-window behaviour (HostSwitcher, in-window probes);
everything else hands off to shell.openExternal so http/https links
open in the user's default browser.
|
||
|
Bohdan Triapitsyn
|
2e5b02e753 |
test: cover external OpenCode proxy routing
- Expose external OpenCode base URL to proxy runtime - Add regression coverage for provider requests through external OpenCode hosts |
||
|
Bohdan Triapitsyn
|
d9fdd39f99 |
fix: restore proxied chat event streaming
- Add dedicated SSE passthrough for proxied chat events - Keep generic API proxy behavior for non-stream requests - Add regression coverage for nginx-safe SSE headers |