Monologue

Monologue Skill

What It Does

The Monologue Skill makes Bea speak spontaneously when no one is talking to her. After a configurable idle timeout, she picks a topic and starts telling an unprompted story, anecdote, or opinion — just like a real streamer filling dead air. The story unfolds in chunks over time, with natural pauses between sections.

File

TERMINAL

src/modules/skills/implementations/monologue.py

State Machine

The skill operates as a two-state machine inside its update() loop:

TERMINAL

         ┌─────────────────────────────┐
         │   IDLE STATE                │
         │   (waiting for timeout)     │
         │                             │
         │  global_idle_time           │
         │  > interval_seconds?        │
         └────────────┬────────────────┘
                      │ YES
                      ▼
         ┌─────────────────────────────┐
         │   STORYTELLING STATE        │
         │   (telling current story)   │◄──────┐
         │                             │       │ continue
         │  time_since_speech          │       │ chunk
         │  > chunk_pause_seconds?     ├───────┘
         └────────────┬────────────────┘
                      │ story ends / closed
                      ▼
               back to IDLE

IDLE mode

Reads the timestamp of the last message in HistoryManager. If now - last_message_time > interval_seconds, triggers _start_new_story().

STORYTELLING mode

After each spoken chunk, waits for chunk_pause_seconds of silence (i.e. time_since_speech > chunk_pause_seconds), then calls _continue_story(). This simulates natural stream-of-consciousness speech with pauses.

Topic Generation

When starting a new story, the skill:

Calls LLM.chat() with the monologue prompt and a list of recent_topics to avoid repetition.
The LLM returns a new topic.
The topic is stored in a deque(maxlen=20) to avoid immediate re-use.

The monologue system prompt (data/prompts/monologue.txt) instructs the AI to:

Never ask questions to the audience
Tell a detailed anecdote or opinion
Use stream-of-consciousness style
Not force-close — conclude naturally

Chunk Continuation

Each chunk is generated by calling brain.generate_response() with a system trigger:

TERMINAL

_continue_story()
    ├─ system_prompt = brain.system_prompt + monologue_rules
    ├─ user_trigger   = "(System: Continue the story about '<topic>'. Remember: NO questions. If finished, add [END].)"
    └─ brain.generate_response(user_trigger, system_prompt=combined_prompt)
            └─ uses the global HistoryManager history for context continuity

The story builds on the main conversation history of the brain — there is no separate per-story history buffer. The story_conversation_history attribute exists in the class but is reset and unused between chunks.

The skill detects a story conclusion by checking for the [END] token in the LLM’s response. When found, the token is stripped before calling perform_output_task() and is_telling_story is set to False.

Configuration

TERMINAL

"monologue": {
  "enabled": false,
  "interval_seconds": 30,
  "chunk_pause_seconds": 4.0,
  "prompt_path": "data/prompts/monologue.txt"
}

Key	Default	Description
`enabled`	`false`	Toggle the skill
`interval_seconds`	`30`	Silence duration before Bea starts monologuing
`chunk_pause_seconds`	`4.0`	Seconds of silence between story chunks before the next chunk is generated
`prompt_path`	`data/prompts/monologue.txt`	Path to the monologue rules prompt

Setting interval_seconds: 1 makes Bea start almost immediately — useful for testing.

Interaction with Brain State

The update() method always checks self.context.is_speaking first. If the brain is already outputting audio, the skill resets last_speech_time and does nothing. This prevents the monologue from starting while Bea is mid-sentence.