Feature Request: Hold-to-Record Voice Mode (Alternative to Push-to-Talk Toggle)

Summary

Add an optional hold-to-record voice input mode as an alternative to the current push-to-talk toggle (Ctrl+B to start, Ctrl+B to stop). In hold-to-record mode, the user holds the record key to capture audio and releases it to stop recording and submit for transcription.

Motivation

The current push-to-talk toggle workflow in CLI/TUI requires two separate key presses for every voice message:

1. Press Ctrl+B → recording starts
2. Speak
3. Press Ctrl+B again → recording stops, audio is transcribed

This is functional but ergonomically suboptimal for short, frequent voice messages — the kind of interaction common in messaging apps (Telegram, Discord voice messages). Users accustomed to mobile-style "press and hold to talk" find the toggle pattern disruptive.

A hold-to-record mode would reduce the interaction to a single press-and-hold gesture, matching user expectations from modern voice interfaces.

Current Behavior

• voice.record_key (default: ctrl+b) is a toggle keybinding
• Implemented via prompt_toolkit @kb.add(key) in cli.py (handle_voice_record)
• TUI uses JSON-RPC voice.toggle via tui_gateway/server.py
• Recording state: _voice_recording boolean + _voice_lock mutex
• Auto-stop on silence is supported (supports_silence_autostop)

Proposed Behavior

Add a new config option voice.record_mode with two values:

CLI (prompt_toolkit)

• Detect KeyPress / KeyRelease events (or use a timer-based heuristic)
• On key press: start recording (same as current)
• On key release: stop recording + trigger transcription
• Must handle edge cases: key held during agent run, key released before recorder initializes, etc.

TUI (Ink/React)

• Add onKeyDown / onKeyUp handlers for the record key
• Mirror CLI behavior: down = start, up = stop + transcribe

Config

voice:
  enabled: true
  record_key: "ctrl+b"      # existing
  record_mode: "toggle"     # new: "toggle" | "hold"

Technical Considerations

prompt_toolkit Limitation

prompt_toolkit's standard @kb.add(key) API only fires on key press, not release. Implementing hold-to-record requires one of:

1. Raw input mode: Use prompt_toolkit.input.vt100_parser or lower-level stdin to capture KeyPress/KeyRelease sequences directly
2. Timer heuristic: On first press, start recording; if no second press within ~500ms, treat as "hold" mode and stop on next release (complex, error-prone)
3. Separate keybinding: Use a modifier key (e.g. ctrl+b = toggle, ctrl+shift+b = hold) — avoids parser changes but adds cognitive load

Option 1 is cleanest but requires testing across terminals (macOS Terminal, iTerm2, VS Code terminal, Windows Terminal).

TUI (Ink)

Ink (React for terminal) supports useInput with isActive flag. Adding onKeyDown/onKeyUp may require:

• Extending the TUI's input handling layer (ui-tui/src/)
• Or using a global key listener with press/release tracking

Backward Compatibility

• Default remains toggle — no breaking change
• record_mode is optional; omitted = toggle
• Both CLI and TUI must respect the setting

Files Likely to Change

Labels

Suggested: type/feature, comp/cli, comp/tui, tool/voice

Priority

P3 — cosmetic, nice to have. Current toggle mode works; this is a UX enhancement.

Submitted by: @vokasug