Reworks how devices connect to an OpenChamber server, end to end.
Pairing v2:
- One-time pairing links/QR codes (openchamber://connect?v=2) carrying a set of transport candidates (LAN/tunnel/relay) and a single-use secret redeemed server-side; no tokens embedded in links
- Add-a-device dialog written for first-time users: intent-based transport choice (Anywhere / Home network only / This computer only) with plain-language descriptions, transparent fallback checkboxes, server-authoritative LAN detection, high-res QR dialog
- Private relay folded into pairing as a transport candidate with a demand-driven lifecycle (enables when a relay device is paired, disables when none remain)
Multi-transport devices:
- A saved device holds all its transports and one token; mobile re-probes on connect, resume, and network change and hot-switches LAN<->relay seamlessly (no re-pairing, no remount, session preserved)
- Desktop can import relay pairing links, switch to relay hosts through the E2EE tunnel, and restore a relay default host after relaunch
Device management:
- Device list (web + desktop) shows live per-device connectivity with the active transport (Connected - Local network / Relay) and platform badges (iOS/Android/macOS/Windows/Linux)
- One physical device = one record: stable per-install dedupe keys across pairing and password re-login; typed pairing label names the device, paired devices name the connection by the issuing server hostname
- Trusted desktop-local client manages all devices (list, revoke, clear revoked); relay host reaps dead client sockets after 3 missed keepalives
Android:
- LAN transport unblocked (cleartext + mixed content, mirroring iOS ATS exceptions); resume re-probe retries through network flux and silently auto-reconnects from a disconnected state
Adds a chat code block wrap toggle in markdown code block headers
Persists and restores the setting across desktop/web settings
Adds localized labels and OpenChamber search entry for the new option
Adds OpenChamber Relay — an opt-in way to reach an instance from a phone,
browser, or another desktop from anywhere, with no open inbound ports, no
tunnel, and no shared LAN. The instance dials outbound to a relay; all app
traffic (HTTP, the event stream, terminal, dictation) is multiplexed and
encrypted through a single connection per client, so the relay only ever
forwards opaque ciphertext.
Transport
- End-to-end-encrypted channel over WebCrypto (ECDH P-256 -> HKDF ->
AES-256-GCM) with a capability-negotiated handshake and a small
HTTP/SSE/WebSocket multiplexing protocol. A byte-compatible JS host mirror
is cross-checked by tests.
- Host: outbound connection manager, per-client tunnel dispatcher to the local
server over loopback, reuse of the existing instance identity key, and
management routes. Disabled by default; explicit opt-in.
- Client: plugs into the existing runtime layer (runtime-fetch/-url/-switch/
-auth, event pipeline, terminal, dictation) so features work over the relay
unchanged; direct-URL and Electron realtime-proxy paths are untouched.
Pairing & UX
- Relay section in Settings -> Remote Instances (live status, QR/link pairing,
revocation via the existing client-token list) and the mobile connect flow.
- Frame batching and idle-gated keepalive keep tunnel message volume low
without affecting streaming smoothness.
Security
- The tunnel is transport only; the server authenticates every tunneled
request exactly as for a direct remote client.
fragments only. The relay stores no keys, tokens, or payloads.
Operability
- The endpoint can be pinned to a self-hosted rel
paired clients inherit it from the offer automatically.
- Relay module DOCUMENTATION.md and a relay-trans
invariants that future WebSocket/streaming changes must follow.
The relay transport is complete and tested; the UI for enabling and pairing
is gated behind openchamber_relay_gate and stays
Adds a server-side "small model" capability: direct, cheap LLM calls that
reuse the user's existing OpenCode provider logins — the mechanism OpenCode
uses internally for titles and summaries but does not expose through the
SDK or plugins. Zero new dependencies; plain fetch with per-provider wire
formats, credentials never leave the server.
Core (packages/web/server/lib/small-model):
- Resolution mirrors OpenCode's session scoping: explicit settings override
→ small_model from the OpenCode config → family scan within the session's
provider → the session's own model. The global provider scan only serves
callers without a session context, and background callers forbid it
entirely (restrictToPreferredProvider), so conversation content never
reaches a provider the user didn't pick — explicit choices excepted.
- Per-provider auth replicating OpenCode's plugin loaders: GitHub Copilot
(device token as bearer, no exchange), ChatGPT plan via the codex
Responses API (single-flight OAuth refresh written back to auth.json),
Anthropic messages, Google generateContent, generic OpenAI-compatible.
- OpenCode's free models (opencode/big-pickle, *-free) are never called
directly; unauthenticated providers are skipped by design.
- Prompt clamping to the model's catalog context limit; thinking disabled
where a wire switch exists (Z.AI/GLM, MiniMax-M3, Gemini Flash); robust
content parsing with a clear error when a thinking model spends its whol
budget on reasoning.
- Settings → Sessions gains a Small Model group: use-default checkbox plus
an override picker limited to authenticated providers, persisted with
web/desktop/VS Code sanitization parity.
Consumers:
- Session assist: a server-side watcher on the global SSE hub generates a
short recap and one suggested follow-up after a session idles quietly fo
a minute, stored on session metadata (openchamber.assist). Freshness is
keyed to the last assistant message id, so new activity invalidates the
payload everywhere with no extra writes. The chat shows the recap under
the last message after five quiet minutes and the suggestion as a
dismissible chip above the composer (tap fills the input, never sends).
Gated by a new Chat setting (default on) that is a hard generation
switch. Language is anchored to the conversation itself, with a
script-mismatch guard against model/backend language hallucination.
- TTS: a third input mode, summarized — long replies are condensed to
spoken prose before playback on any TTS engine.
- Git: commit-message and PR generation moved off the active chat session
onto the small model fed with real diffs and the commit list (bodies
included), with a session-transport fallback for free-model-only setups.
- Notes: Add to notes distills long selections into 1-3 dense sentences
preserving exact identifiers, with verbatim fallback on failure.
Fixes along the way:
- The global event watcher now starts unconditionally; it was gated behind
the desktop-notify env, leaving the server-side event hub dead in
packaged apps.
- OpenCode re-emits message.updated for old user messages after idle; the
watcher no longer mistakes those for new activity.
- Session metadata merges from a fresh read right before the PATCH, so
writes made during the generation window (suggestion dismissals, review
links) are preserved; the assist runtime stops during graceful shutdown.
Complete rebuild of voice input on a server-authoritative streaming
architecture, replacing the legacy Web Speech / whole-blob / WASM engines
and the dead voice-agent layer (~4k lines removed).
Speech-to-text (dictation):
- Client streams 16 kHz mono PCM16 chunks over /api/dictation/ws with
seq/ack ordering; buffered audio is retained and replayed on reconnect
- Server transcribes and streams live partial transcripts back;
segments auto-commit every ~15s with silence suppression and adaptive
finalization timeouts
- Local provider (default, zero config): sherpa-onnx models in a forked
worker process — auto-download with progress, staged extraction with
verification, corrupt-model auto-recovery, idle shutdown after 5 min
- Model catalog with settings picker (accuracy/speed ratings, sizes,
download/delete): Parakeet TDT v2 (English) and v3 (25 European
languages, auto-detected), Whisper base and tiny (multilingual, light)
- OpenAI-compatible provider for any Whisper endpoint
- Composer overlay with live transcript, volume meter, timer, and
cancel / insert / insert-and-send actions; failed transcriptions keep
their audio for retry or accepting the partial text as-is
- Configurable keyboard shortcut (default mod+alt+v) toggles dictation;
Enter confirms and Escape cancels while recording
- Overlay is pixel-aligned with the composer (measured footer height,
matching paddings/typography/gaps) — no layout shift when toggling
Text-to-speech:
- Local Kokoro provider (English, 11 voices) synthesized in the same
worker via /api/dictation/tts/speak, managed by the shared model
pipeline; sentence-pipelined playback keeps time-to-first-audio at
~1 sentence regardless of message length, and stop cancels in-flight
synthesis
- Sanitizer keeps inline-code content (strips backticks only), reads
interword slashes aloud, and removes only absolute file paths
Settings:
- Voice page unified: a single read-aloud toggle owns all playback
options (the confusing "Enable Voice Mode" is gone); a new "Enable
voice input" toggle (default on, persisted to settings.json) hides
the composer mic entirely when disabled
Mobile and transport:
- iOS/Android microphone permissions added (dictation was previously
impossible on mobile)
- Fixed Android WebSocket upgrades: the Capacitor WebView origin
(https://localhost) was missing from the packaged-client allowlist,
403-ing every WS connection — root cause of the old mobile SSE lock,
which is now removed for all transports
Security and conventions:
- All HTTP routes sit behind the global /api auth gate; the WS upgrade
explicitly validates the UI session and origin, with oc_url_token
narrowly allowlisted and covered by tests; the dictation socket mints
a fresh URL token before connecting
- Routes register before the generic OpenCode proxy; the client goes
through runtimeFetch/getRuntimeUrlResolver, and runtime switches
reset the dictation socket
- VS Code deliberately reports dictation as unavailable (no server
process in that runtime)
CI: workflow Node bumped 20 -> 22 to match the repo engines and fix
better-sqlite3 installs broken by node-gyp@latest on Node 20.
New dependency: sherpa-onnx-node (prebuilt N-API; macOS/Linux x64+arm64,
Windows x64 — Windows-on-ARM falls back to the OpenAI-compatible provider)
* feat: support OpenCode steer delivery / follow-up behavior settings
Implements issue #1766 — steer delivery mode for mid-turn message
insertion, replacing the old boolean queue-mode toggle with a tri-state
follow-up behavior setting (Steer / Queue / Send immediately).
- Plumbing: threaded optional delivery: 'steer' through sendMessage
-> routeMessage -> opencodeClient.sendMessage -> promptAsync
- Store: messageQueueStore stores followUpBehavior; migration from
legacy queueModeEnabled persisted state
- Settings: Chat -> Follow-up behavior shows three radio options
using existing settings UI patterns
- Composer: when session is busy, a floating queue button remains;
force-sending a queued message (via chip click) uses delivery: 'steer'
during a busy session; Steer button intentionally omitted — steer is
available via the two-gesture path (Enter to queue -> chip to steer)
- Keyboard: queue mode = Enter queues, Ctrl+Enter sends; otherwise
Enter sends, Ctrl+Enter queues
- Persistence: DesktopSettings, web settings payload, and server-side
sanitizer handle the new key with legacy fallback
- i18n: follow-up behavior section and option labels in all 9 locales
plus new chat.chatInput.actions.queue label
- Search: settings registry updated from chat.queue-mode to
chat.follow-up-behavior
Validation: type-check passes (no new errors), lint clean.
* fix(#1766): make steer mode actually steer
The followUpBehavior === 'steer' branch in handlePrimaryAction and the
keyboard handler was a no-op — both fell into the else branch and sent
without the delivery: 'steer' flag, so selecting 'Steer (insert into
the running turn)' in settings produced identical behavior to 'Send
immediately'.
- handlePrimaryAction: when steer mode is selected and the session is
busy, call handleSubmit({ delivery: 'steer' }) directly
- Keyboard handler: in steer mode, Enter steers and Ctrl+Enter sends
immediately (consistent with queue mode where Ctrl+Enter bypasses
the special handling)
Also removes the unused chat.chatInput.actions.queue i18n key from all
9 locales (it was a dead key after the Steer button was removed from
the composer).
Validation: type-check clean, lint clean.
* refactor(#1766): flatten nested ternary in followUpBehavior resolution
Replace nested ternary with explicit if/else chain per project code style
(CONTRIBUTING.md). Import FollowUpBehavior type explicitly for the new
let declaration.
* feat(chat): drop redundant 'immediate' follow-up mode, keep Queue + Steer
'Immediate' was wire-identical to 'Steer' on a busy session: OpenCode only
supports delivery 'steer' | 'queue' and defaults to 'steer', so an immediate
send (no delivery flag) already steered into the running turn. The three-mode
UI therefore exposed two settings that did the same thing.
Collapse to two modes — Queue (unchanged: client-side queue with edit/reorder)
and Steer. Any persisted/legacy 'immediate' (and legacy queueModeEnabled=false)
now maps to 'steer', preserving prior behavior. Removes the immediate option,
its keyboard branch, the i18n label across all locales, and narrows the
followUpBehavior union to 'steer' | 'queue'.
---------
Co-authored-by: Leonid Skorobogatyy <bash@opencode.itc.local>
Co-authored-by: Bohdan Triapitsyn <artmore@protonmail.com>
Show a count of chats (root sessions) with unseen activity on the macOS
dock icon. The count is computed in the existing tray snapshot (full
cross-project list, not the capped tray view; a subtask's unseen rolls up
to its root only when subtask notifications are enabled) and pushed to the
main process over the existing desktop_tray_update IPC, which calls
app.setBadgeCount (0 clears it). The badge clears as sessions are marked
seen on window focus.
Add a Dock badge toggle in Appearance settings (default on, persisted,
darwin desktop only), localized across all dictionaries, with a matching
settings-search entry whose availability mirrors the render guard exactly.
* feat(agents): expose thinking variant configuration in agent settings
Fix#1425: add variant field to agent config UI so users can configure
thinking/reasoning depth per agent without editing opencode.json.
Changes:
- Added variant to AgentConfig and AgentDraft types in useAgentsStore
- Pass variant in createAgent and updateAgent API calls
- Support null for temperature, top_p, and variant to clear overrides
- Added variant input field in AgentsPage 'Model & Parameters' section
- Added variant to settings search registry
- Added i18n strings for variant field in all 8 non-English locales
The variant field maps to provider-specific parameters (e.g. Anthropic
high/max variant, OpenAI reasoning effort). Users can enter any string
value; the SDK passes it through to the model provider.
Clearing temperature/topP/variant now sends null to the server instead
of omitting the field, which properly removes the override in
opencode.json.
* fix(sync): preserve tool state.time in materialization merge
* chore: trigger re-review
* fix(agents): use thinking variant selector in settings
* fix(agents): preserve thinking variant values
---------
Co-authored-by: Leonid Skorobogatyy <bash@opencode.itc.local>
Adds item-level search inside Settings so users can find concrete settings like provider auth, agent mode, terminal font size, tunnel options, notification events, and similar controls instead of only filtering top-level pages.
Groups search results by Settings page and shows localized labels plus optional descriptions where useful.
Supports keyboard navigation with Arrow Up/Down, Enter, and Escape, matching the existing autocomplete interaction style.
Opens the correct Settings page or split-page draft state before scrolling to the matching control.
Highlights the matched setting with a subtle token-based background so users can see where they landed without an aggressive outline.
Adds explicit data-settings-item anchors across Settings pages and a centralized search registry with runtime/mobile availability guards.
Updates Settings UI skill guidance so future Settings changes keep search registry entries, anchors, localization, and availability guards in sync.
Iuliia Ivashko