Turtle
38c7a9fbf4
* feat(release): publish package source, declaration maps, and documentation Every release member publishes its TypeScript source and the declaration maps that resolve into it, and @deepseek-ai/dsh publishes the repository Markdown documentation staged from docs/ at prepack time. The ./src/* export most members already declared now resolves for an installed consumer, and a consumer editor follows a declaration to the implementation. Bundled JavaScript maps stay unpublished; the vendored family and the Desktop runtime file policy are unchanged. Rationale: .agents/notes/implemented/process/2026-10-08-published-source-and-docs.md * feat(release): publish JavaScript maps and compress bundles Every emitted JavaScript map ships beside the bundle it describes, and the workspace build and the client preset compress bundles without mangling identifiers. The documentation payload now mirrors the website publication manifest and is removed again by `postpack`; Desktop and the Python executable exclude the published source, documentation, and maps; `clean` removes a staged copy. Typert's publication contract gains the Remote declaration map. * fix(release): keep node bundles readable for the worker image The browser worker image lowers `yield*` only as a whole statement and rewrites `for await` loops by their source layout, so compressed node-side entries either fail that lowering or lower to unparseable code. Only the browser bundles keep compression, which removes about a third of their text. The bundle-purity assertion now matches the module-loader call regardless of the quote style compression picks, and the release-pack fixture exercises the installed package's new documentation requirement. * feat(release): publish the whole documentation tree `@deepseek-ai/dsh` copies `docs/` in full instead of the website's page set: the maintainer documents, the generated migration records, the translation guidelines, and the `*.i18n.yaml` consistency records ship with the installed package, so the documentation's own links resolve without a checkout. * feat(release): publish the upgrade guides with the documentation The staged documentation keeps the website's page set and adds every `docs/upgrade-guide/` document, which is what an installed reader consults when a released change breaks something. The rest of the maintainer tree stays out. * test(release): expect source and maps in master's new payload cases * feat(release): keep installed docs and source readable in Desktop Desktop keeps the published src/ and docs/ trees inside app.asar, and every model shell call receives DSH_INSTALL_DIR and DSH_NODE_EXECUTABLE so an agent can read them through Electron's Node mode. Staging also generates docs/upgrade-guide/README.md, an index of the upgrade guides grouped by the release they upgrade from. Minified client bundles print string literals with backticks: the worker image packer now recognizes those chunk requests, client license banners are prepended after minification so their text survives, and the affected artifact tests accept either quote. * test(desktop): read archived docs through the Windows PowerShell form Regenerate the config catalog for the shifted shell-env source line. * test(desktop): wait for the GUI-subsystem Electron process from PowerShell * build(release): publish source and maps from the message-quotes package * build(release): publish source and maps from packages new on master * build(release): publish source and maps from the tool-office package
description, kind
| description | kind |
|---|---|
| Record speech and insert reviewable transcripts into the conversation draft. | package-reference |
@deepseek-ai/dsh-experimental-client-ui-voice-input
English | 中文
Summary
This optional browser plugin adds an outline microphone icon between the model selector and Send. When recognition is ready, clicking it opens a recording toolbar with measured audio levels, Cancel and Stop. The waveform stays shorter than the recording buttons. Stopping transcribes into the draft. Recognition preferences and model preparation live in plugin settings; language choices come from the selected provider.
Table of Contents
- Use this package
- Understand the implementation
- Further Exploration
- Model Experience
- Known Limitations and Deferred Work
- Dev Note
Use this package
Enable the voice input bundle from Plugins. After cache inspection, missing models trigger a setup prompt with Go to setup and Later actions. Go to setup opens bundle details; Download and prepare starts installation there. Intact cached models need no prompt or repeated installation. Before recognition is ready, guidance appears only on click, with no microphone tooltip. Clicking the microphone opens the same guidance dialog; its action opens the voice bundle details without recording or downloading. Cache inspection, preparation in progress, failures and disconnected services instead offer navigation to status, progress or error details. The dialog moves focus to its navigation action and keeps Tab traversal inside it; Later or Escape restores the previous focus without changing or submitting the draft. Before preparation, local providers can show their estimated disk, memory and setup-time requirements. The preparation summary starts collapsed; expand it to inspect completed, current and pending steps. Once ready, click the microphone, allow access and click Stop to transcribe. Cancel or Escape discards capture. Recognition feedback stays inside the toolbar. Provider and language preferences are saved from bundle details through the Settings service. Browser microphone access requires HTTPS or loopback and operating-system permission.
Use the native Input device selector in bundle details to follow the system default or select a browser microphone. System default shows the current device name when available. While bundle details are active, the selected input's live level remains visible at the right of the field; selecting another input restarts the preview for that device. Hiding or leaving the page, or disabling the plugin, releases the preview, including late permission grants. The browser remembers the device for subsequent recordings independently of Host recognition settings. Device insertion and removal refresh the options without interrupting an available selected input. A disconnected selection retains its name and position as a disabled option with an Unavailable label. Reconnecting while the page is active restores its preview; choosing another input removes the unavailable option. Interrupted recordings require an explicit restart. Permission denial keeps other choices accessible; an explicitly selected device never silently falls back to another input.
Before downloading or retrying, Model download source offers Automatic and the origins advertised by the Host. SenseVoice normally offers Hugging Face and HF-Mirror; a fixed private deployment exposes only its configured origin. Manual selection uses only that source, while Automatic preserves the provider’s comparison and fallback policy. The choice is retained for retries in the current card, without changing deployment configuration or recognition preferences. Source controls are unavailable during preparation and hidden when resources are ready.
pluginNavigation is optional: the plugin starts without it, and the setup dialog offers the details button only while the service is present.
Understand the implementation
Maintainer details — click to expand
Native MediaRecorder captures audio and Web Audio converts it to the Host PCM format after flushing the final recorder chunks. Input-level previews use Web Audio without creating a MediaRecorder. Window blur during the microphone permission request does not cancel capture; blur during recording does. Capture failures immediately end the recording UI and offer retry inside the toolbar. Cancellation, failure and plugin disposal share one resource-release promise; the plugin retains ownership until AudioContext closure settles. The original editor selection carries a draft revision: changed drafts retain the transcript for explicit insertion or discard beside the microphone. Session changes, hiding the page during capture and plugin disposal invalidate late results and release tracks. One plugin-owned readiness subscription serves the composer, installation prompt and details. These views use the existing Slot lifecycle and own no preparation tasks.
Further Exploration
Model Experience
None, as recordings and preparation remain outside model requests; ordinary user submission owns any later text.
KV Cache effect
No direct effect; ordinary submission owns the message content.
Known Limitations and Deferred Work
- No automatic send, always-on microphone, wake word, streaming captions or speech synthesis. “Local” means the Host machine, which can differ from the browser’s machine.
Dev Note
Maintainer details — click to expand
None.