feat(chat): add voice input with real-time transcription

[guitavano]

Summary

• Adds a microphone button to the chat input (only shown when the browser supports SpeechRecognition)
• Real-time speech-to-text using the browser-native Web Speech API — no backend changes required
• Live waveform visualizer driven by the Web Audio API (AnalyserNode) animates while listening
• Transcribed text is appended to any existing editor content (not replaced), so the user can mix typing and dictation freely
• Cancel (×) and confirm (✓) controls match the existing input style
• Proper microphone permission handling: getUserMedia is called upfront to trigger the browser prompt; if denied, the mic button turns red with a descriptive tooltip and clicking it re-prompts

Test plan

• Click the mic button — browser should ask for microphone permission
• Deny permission — mic button should turn red with tooltip "Microphone access denied — click to try again"
• Grant permission — waveform overlay should appear and animate while speaking
• Speak — interim transcript (dimmed) should appear live, final transcript (solid) should accumulate
• Press ✓ — transcribed text should be inserted at the cursor / appended to existing text in the editor
• Press × — overlay should dismiss with no changes to the editor
• Type something first, then use voice — transcribed text should be appended after existing content
• Test on Firefox (Speech API unsupported) — mic button should not appear

Made with Cursor

---

Summary by cubic

Add voice input to chat with real-time transcription that types directly into the editor and shows a live waveform. During recording the editor locks; cancel restores previous text, confirm keeps the transcription.

• New Features

  • Mic button shows only when SpeechRecognition is available; hidden on unsupported browsers.
  • Transcript appears live in the editor while recording; input is disabled to prevent edits.
  • Bottom bar switches to waveform with ×/✓; cancel restores prior content, confirm keeps the text.
  • Permission flow via getUserMedia; denied state turns the mic red with a tooltip and allows retry.
  • New use-voice-input hook and VoiceWaveform; TiptapInput adds appendText(), syncVoiceText(), and restoreContent().

• Bug Fixes

  • Added Web Speech API global types in globals.d.ts for SpeechRecognition and related interfaces to fix TypeScript builds and noUncheckedIndexedAccess issues.
  • Prevented losing last interim words on stop and ensured a space is added before appended voice text when baseline content exists.
  • Removed unused VoiceInputOverlay export.

^{Written for commit ec23fa4. Summary will update on new commits.}

[github-actions]

🧪 Benchmark

Should we run the Virtual MCP strategy benchmark for this PR?

React with 👍 to run the benchmark.

Reaction	Action
👍	Run quick benchmark (10 & 128 tools)

Benchmark will run on the next push after you react.

[github-actions]

Release Options

Suggested: Minor (2.261.0) — based on feat: prefix

React with an emoji to override the release type:

Reaction	Type	Next Version
👍	Prerelease	2.260.3-alpha.1
🎉	Patch	2.260.3
❤️	Minor	2.261.0
🚀	Major	3.0.0

Current version: 2.260.2

Note: If multiple reactions exist, the smallest bump wins. If no reactions, the suggested bump is used (default: patch).

[cubic-dev-ai]

1 issue found across 4 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="apps/mesh/src/web/components/chat/input.tsx">

<violation number="1" location="apps/mesh/src/web/components/chat/input.tsx:334">
P1: Transcription can be lost on confirm because `appendText` is called while `TiptapInput` is unmounted in recording mode.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[cubic-dev-ai]

2 issues found across 3 files (changes from recent commits).

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="apps/mesh/src/web/components/chat/input.tsx">

<violation number="1" location="apps/mesh/src/web/components/chat/input.tsx:337">
P1: Use the final text returned by `stopRecording()` when confirming, otherwise the last dictated words can be lost.</violation>

<violation number="2" location="apps/mesh/src/web/components/chat/input.tsx:354">
P2: Prefix dictated text with a space when baseline content is non-empty so transcription truly appends instead of concatenating words.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

[viktormarinho]

lgtm