Configuration Reference

The system is configured via config.yaml in the root directory. All services share this file, allowing for centralized management.

Service Ports

Service	Port	Description
Orchestrator	8000	Main coordinator and Web UI
STT Service	8001	Speech-to-Text WebSocket API
LLM Service	8002	Language Model API
TTS Service	8003	Text-to-Speech WebSocket API

Core Configuration

Orchestrator

orchestrator:
  host: "0.0.0.0"
  port: 8000
  stt_websocket_path: "/ws/stt"
  log_level: "INFO"
  enable_latency_tracking: true
  system_prompt_file: "orchestrator/system_prompt.txt"
  hotkeys:
    toggle_listening: "ctrl+shift+l"
    cancel_speech: "ctrl+shift+c"

LLM Providers

Supported: ollama, gemini, llamacpp.

llm:
  provider: "ollama"  # Options: gemini, ollama
  providers:
    gemini:
      model: "gemini-2.5-flash"
      api_key: ""  # Optional: Override GEMINI_API_KEY env var
      generation_config:
        thinking_budget: 0  # 0 = disabled, -1 = dynamic (Gemini 2.5), 2.5 pro 128-32768, 2.5 flash 0-24576
        # thinking_level: "low"  # Gemini 3 Pro low/high, Flash low/high/medium/minimal
        temperature: 1.0
        top_p: 0.95
        top_k: 40
        max_output_tokens: 8192
    ollama:
      model: "Qwen3-8B-Q4-8kcontext"
      base_url: "http://localhost:11434"
      timeout: 300.0
      disable_thinking: true
      generation_config:
        temperature: 0.7
        top_p: 0.9
        top_k: 40
        num_predict: 2048
    llamacpp:
      model_path: ""
      n_ctx: 4096
      n_threads: 0
      n_gpu_layers: -1

STT Configuration

Supported: funasr (recommended), faster-whisper.

stt:
  host: "0.0.0.0"
  port: 8001
  provider: "funasr"
  language_code: "zh"
  sample_rate: 16000
  interim_transcript_min_samples: 16000
  providers:
    faster-whisper:
      model_path: "faster-whisper-small"
      device: null  # null = auto-detect
      compute_type: null  # null = auto-detect
    funasr:
      model_name: "paraformer-zh-streaming"
      vad_model: "fsmn-vad"
      punc_model: "ct-punc"  # Optional punctuation model
      vad_kwargs:
        max_single_segment_time: 30000
      device: null
      batch_size_s: 0
      streaming:
        enabled: true
        chunk_size: [0, 8, 4]
        encoder_chunk_look_back: 4
        decoder_chunk_look_back: 1
        vad_chunk_size_ms: 100
        silence_threshold_ms: 300

TTS Configuration

Supported: genie-tts, edge-tts, chattts, elevenlabs, gpt-sovits.

tts:
  host: "0.0.0.0"
  port: 8003
  provider: "genie-tts"
  providers:
    edge-tts:
      voice: "zh-CN-XiaoyiNeural"
      rate: "+0%"
      pitch: "+0Hz"
    chattts:
      model_source: "local"  # Options: local, huggingface, custom
      device: null
    elevenlabs:
      voice_id: ""
      stability: 0.5
      similarity_boost: 0.8
      style: 0.0
    genie-tts:
      character_name: "ema"
      onnx_model_dir: "path/to/model"
      language: "jp"  # Options: zh, en, jp
      reference_audio_path: ""
      reference_audio_text: ""
      source_sample_rate: 32000
    gpt-sovits:
      server_url: "http://127.0.0.1:9880"
      default_reference: "default"
      default_text_lang: "zh"  # Default language for TTS input text
      streaming_mode: 2  # 0=off, 1=best quality, 2=medium, 3=fastest
      temperature: 1.0
      top_p: 1.0
      top_k: 15
      speed_factor: 1.0
      timeout: 30.0
      references:
        default:
          ref_audio_path: "references/default.wav"
          prompt_text: "这是一个默认的参考音频"
          prompt_lang: "zh"
        character1:
          ref_audio_path: "references/character1.wav"
          prompt_text: "我是角色1，说话温柔优雅"
          prompt_lang: "zh"

Audio Configuration

audio:
  input:
    sample_rate: 16000
    channels: 1
    device: null  # null = default device
  output:
    sample_rate: 32000
    channels: 1
    device: null  # null = default device
  dtype: "float32"
  block_size_ms: 100
  silence_threshold_ms: 500  # Used as fallback for STT providers
  listening_status_poll_interval: 1.0

OBS Configuration

obs:
  websocket:
    host: "localhost"
    port: 4455
    password: ""
  subtitle_source: "subtitle"
  subtitle_ttl_seconds: 10
  visibility_source: ""  # Optional
  appear_filter_name: ""  # Optional
  clear_filter_name: ""  # Optional

Bilibili Configuration

bilibili:
  enabled: true
  room_id: 31232063
  sessdata: ""  # Optional SESSDATA cookie
  danmaku_ttl_seconds: 60

Service URLs

Configuration for inter-service communication (used by orchestrator).

services:
  orchestrator_base_url: "http://localhost:8000"
  stt_websocket_url: "ws://localhost:8001/ws/transcribe"
  tts_websocket_url: "ws://localhost:8003/synthesize/stream"
  ocr_websocket_url: "ws://localhost:8004/monitor/stream"
  ocr_base_url: "http://localhost:8004"

Hot-Reloading

The Orchestrator monitors the system_prompt_file for changes. Updating the text file will automatically update the agent's persona without restarting.