feat(dashboard): add dictation and live voice API controls

[lidge-jun]

Summary

• Add separate Dictation and Live Voice sections under Connections/API keys, with configured versus observed state, ephemeral data-key input, bounded file upload/cancel and native session readiness checks.
• Publish validated audio endpoint metadata through the existing management response. Keep audio calls outside management credential injection, and keep secrets out of caches, examples and event displays.
• Ordinary dependent stack: feat(audio): expose standalone transcription API #4391 (file transcription) -> feat(audio): expose streaming dictation and owned live calls #4392 (streaming and call ownership) -> this PR (dashboard). Parent branch is codex/audio-streaming; final integration target remains dev. No native GitHub stack or merge requested.

Verification

• Local product tests, typecheck, build and install: NOT RUN, explicitly prohibited by the owner. Commit and push use --no-verify.
• Final head 066091134b5d30297fb9b9c4460378a689ecb41f: remote gates 103544812802 passed lint, typecheck, 1,998 dashboard tests / 0 failures, privacy and build. Screenshot/description gate also passed. Runtime/UI source is unchanged from f5aefd88af3116bec4f2ddc72c0c6fa974f52a83, whose remote metadata tests passed 10/10.
• Actual CI-built dashboard, synthetic responses only: upload/copy, HTTP 401, cancellation, native session readiness/disconnect, inactive-tab cleanup and no key persistence passed. No page JavaScript errors. Desktop/tablet/mobile widths 1440/1024/768/390/320 and Korean narrow labels inspected. Artifact merge bb0f30fbab4496b2b69bbc1e8148e59035ba3778 GUI tree exactly equals the runtime/UI source tree.
• Independent security/component source reviews closed all actionable findings, including metadata extras, post-readiness errors, replacement races, idle status and mobile scroll-spy. Final delegated review used xai/grok-4.6 as requested.

• Parent head 011f2dff5ca3667b88f090fe711c4b2c77efd190 passed all new audio cases and gates in run 34687731903. Existing journal restore failures remain separately recorded per owner scope; no whole-suite-green claim.
• No personal recordings, real upstream audio requests, account changes, service restart or deployment used for QA.
• Final-head test 2/4 has only the separately scoped journal restore failures: hashless native restore refusal and damaged managed-default fallback cleanup. Other platform shards may still be completing. No whole-run-green claim; leave this PR draft and unmerged.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed.
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

✅ Deterministic PR hygiene checks passed.

[github-actions]

✅ READY

• all PR quality gates passed.

Hygiene

✅ Deterministic PR hygiene checks passed.

[lidge-jun]

리뷰 · 우선순위 50 / 80

이 PR은 오디오 스택 wp3(대시보드)입니다. 부모는 #4392(codex/audio-streaming)이고, Connections/API keys에 Dictation·Live Voice 구역을 나란히 둡니다. 지금 dev HEAD 20861aebf의 GUI에는 이 패널이 없고, management buildApiAccessEndpoints도 오디오 메타데이터를 주지 않습니다. 이 브랜치는 gui/src/components/apikeys-workspace/AudioApiPanel.tsx, gui/src/audio-api-client.ts, audio-api-examples.ts, i18n, 스타일을 추가하고 src/server/management/api-access.ts에 audio 블록(엔드포인트·모델·configured 플래그)을 붙입니다. 관리 API 응답으로 URL/설정 여부만 공개하고, 실제 오디오 호출에는 management 자격증명을 주입하지 않으며, 키는 입력창 ephemeral 상태로만 둡니다(캐시·예제·이벤트에 시크릿 없음).

dev 본선 작업(#4390 등)과 직접 충돌하지 않지만, #4391 → #4392 → 이 PR 의존이라 부모 두 개가 모두 draft·CI 미완료면 여기도 머지할 자리가 없습니다. PR 본문도 스크린샷 붙이기 전엔 draft를 유지하고, 로컬 제품 테스트는 금지·원격 CI만 게이트라고 적었습니다. 이 시각 기준 상위 스택 CI에 test 2/4 실패가 남아 있고, 이 PR 자체 hygiene/label 등은 재실행 중입니다. GUI 회귀 테스트(audio-api-panel, audio-api-client, auth-memory)는 추가돼 있습니다.

UI 역할 분리는 분명합니다. DictationPanel은 파일 업로드+취소+트랜스크립트 표시와 스트리밍 예제(복붙)를 보여주고, LiveVoicePanel은 WebSocket 연결 상태·이벤트 타입 몇 줄을 보여 줍니다. “configured”는 계정 health/ entitlement가 아니라 프로바이더 설정 존재만 뜻한다고 api-access.ts 주석에 박혀 있어, 초록 표시만 보고 받아쓰기가 된다고 오해할 여지는 문서로 더 막아야 합니다. 스트리밍 받아쓰기는 예제 코드만 있고 패널에서 직접 마이크 세션을 열지는 않습니다(의도적 범위로 보임).

src/server/management/api-access.ts AudioApiAccess - transcriptionConfigured/dictationConfigured/liveConfigured가 설정 존재만 본다. GUI 배지 문구(audio.configured)가 “준비됨”처럼 들리면 사용자 오해가 난다. “설정됨(계정 검증 별도)” 쪽으로 i18n을 다듬자.
gui/src/components/apikeys-workspace/AudioApiPanel.tsx DictationPanel - 파일 선택 accept에 audio/*,.mp4,.webm만 있다. 서버가 받는 포맷과 다르면 업로드 후 400만 난다. 서버 allow-list와 맞추거나 힌트 문장을 달자.
gui/src/audio-api-client.ts connectLiveAudio - 브라우저가 마이크/SDP를 어떻게 준비하는지가 패널 UX에 안 보인다. 지금은 “연결”만 있어, 실제 통화 성립 여부를 기여자가 스크린샷으로 검증하기 어렵다. draft 해제 전 데스크톱·모바일 캡처가 필수인 이유다.
의존 스택 #4391/#4392 - 부모 CI가 빨간 동안 이 PR만 고쳐도 통합 검증이 안 끝난다. 스크린샷·i18n만 먼저 닫지 말고 부모 exact-head 녹색을 기다리자.
docs-site proxy-formats 다국어 - 오디오 엔드포인트 문서가 여러 로케일에 퍼진다. 영문 기준 문장과 번역이 어긋나면 로케일 패리티 테스트에 걸릴 수 있으니 한 줄씩 맞추자.

메인테이너의 판단이 필요한 지점

• wp3를 스택 통째로 리뷰할지, wp1/wp2가 dev에 들어간 뒤 UI만 리베이스할지
• 대시보드에 마이크/스트리밍 받아쓰기 실조작 UI가 이번 범위인지, 예제+파일 업로드만으로 충분한지
• configured 배지를 설정/관측/엔타이틀먼트로 더 쪼갤지
• draft 해제 조건으로 스크린샷 종류(데스크톱/모바일/다크)를 어디까지 요구할지

너의 추천
이 PR은 대기가 맞다. #4391·#4392 CI와 계약을 먼저 고정하고, configured 문구·accept 힌트·실기 스크린샷을 채운 다음에 review-ready로 올리자. 지금은 merge하지 말고 스택 아래부터 녹색을 만들자.

이 댓글은 grok-bot이 작성했습니다

[lidge-jun]

Maintainer integration record — audio stack tip, merged by @lidge-jun.

This is the tip of the cumulative audio stack, so merging it lands the whole lane in one commit: #4391 (transcription API), #4392 (streaming dictation and owned live calls), and #4395 (dashboard controls). #4391 was squashed separately beforehand; #4392's content arrives here.

• Exact head at integration: ec9e3734e80a32e2b60d7ace89d23de94fae0123.
• CI evidence: Cross-platform CI run https://github.com/lidge-jun/opencodex/actions/runs/34731037202 concluded success on this head — Linux test shards 1-4, macOS 1/2 and 2/2, gates, keyring, docker smoke, npm-global smoke, storage policy, api usage.
• Non-tip links carried [skip ci] by design. Tip-only CI for this batch is an explicit repository-owner decision; MAINTAINERS.md otherwise requires a per-PR check, so this is a recorded deviation rather than something the maintainer-integration clause grants on its own. The stack being cumulative is what makes the tip run cover the links beneath it.
• Review state: no outstanding maintainer change requests.
• Security review: not applicable — no authentication, credential, OAuth, workflow, release-automation, or dependency-installation surface.
• Local verification on the lane: typecheck, structure:check, privacy:scan passed; focused tests passed per link. No local full-suite run was used as evidence.

Squashing with --match-head-commit pinned to the head above and a squash message free of [skip ci], so the dev-branch regression run still fires.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-13T02:10:10.162566Z	ec9e373	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ec9e3734e8

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Route connected audio controls to the hub data plane

When a connected-client dashboard uses direct management transport to a wildcard-bound hub, the cross-origin /api/keys request carries the machine dashboard's Origin, which resolveApiAccessBaseUrl prefers over the hub request URL. These new endpoints therefore point to the local machine listener, but src/client/machine-listener.ts rejects every /v1 request and all WebSocket upgrades, so transcription returns 404 and Live Voice cannot connect. Derive executable audio endpoints from the connected hub's inference origin, or provide a data-plane relay instead of reusing this request-origin display base.

Useful? React with 👍 / 👎.