Read Aloud Plugin

The Read Aloud plugin adds a "read aloud" action to each assistant message. Clicking it sends the final message text to the shared speech (text-to-speech) service and plays the audio in the browser. It is a play-on-demand action, not a real-time voice mode.

Configuration

plugins:
  - module: codumentor.plugins.read_aloud
    class: ReadAloudPlugin
    # args:
    #   enabled: true   # (default)

Parameters

• enabled (optional): Whether the plugin is active and exposes the action (default: true)

The plugin has no other settings of its own — it relies on the top-level shared speech: configuration block (the same one the Telegram voice replies use):

speech:
  enabled: true
  base_url: "https://api.openai.com/v1"
  api_key: "${OPENAI_API_KEY}"
  tts_model: "tts-1"
  tts_voice: "alloy"

How It Works

The plugin adds a custom x-read-aloud-action element to the message-actions row (next to the copy button) and exposes a small HTTP API under /ui/plugins/read_aloud:

• GET /status — reports whether TTS is available (requires speech.enabled plus a resolvable base_url and api_key). The button renders disabled with a tooltip when it isn't.
• POST /speak — synthesises the supplied text (as MP3, for browser <audio> playback) via the shared speech service.

There are no agent tools and no turn hooks — the action is driven entirely from the UI.

Notes

• Requirements: speech.enabled: true and an api_key that resolves to a real value via an env var. A per-user ${secret:…} key cannot resolve for this call (it runs outside an agent turn), so the action stays disabled in that case.
• Only the final assistant message is read — never intermediate tool-call narration.
• One message plays at a time; starting another stops the previous one.
• The speech service is shared, so changes to the speech: block affect read-aloud and Telegram voice together.