providers/api.yaml

# ─────────────────────────────────────────────
# OpenHer — Unified API Configuration
# ─────────────────────────────────────────────
# All external API settings in one place.
# ENV var overrides always take priority over this file.
# Future APIs (vision, STT, etc.) add new sections here.

# ── LLM (对话引擎) ──
llm:
  provider: "gemini"             # Active provider name
  # model is auto-resolved from providers.<name>.default_model
  # Override globally: set DEFAULT_MODEL env var
  temperature: 0.92
  max_tokens: 1024
  providers:
    dashscope:
      base_url: "https://dashscope.aliyuncs.com/compatible-mode/v1"
      api_key_env: "DASHSCOPE_API_KEY"
      default_model: "qwen-flash-2025-07-28"
    openai:
      base_url: "https://api.openai.com/v1"
      api_key_env: "OPENAI_API_KEY"
      default_model: "gpt-4o"
    moonshot:
      base_url: "https://api.moonshot.cn/v1"
      api_key_env: "MOONSHOT_API_KEY"
      default_model: "moonshot-v1-auto"
    ollama:
      base_url: "http://localhost:11434/v1"
      api_key_env: ""
      default_model: "qwen3.5:9b"
      no_key_required: true
    gemini:
      api_key_env: "GEMINI_API_KEY"
      default_model: "gemini-3.1-flash-lite-preview"
    stepfun:
      base_url: "https://api.stepfun.com/v1"
      api_key_env: "STEPFUN_API_KEY"
      default_model: "step-3.5-flash"
    minimax:
      base_url: "https://api.minimax.io/v1"
      api_key_env: "MINIMAX_LLM_API_KEY"
      default_model: "MiniMax-M2.7"
    litertlm:
      api_key_env: ""
      default_model: "gemma-4-e2b"
      no_key_required: true

# ── TTS (语音合成) ──
tts:
  provider: "dashscope"            # openai / dashscope / minimax
  cache_dir: ".cache/tts"        # Relative to project root
  providers:
    openai:
      api_key_env: "OPENAI_API_KEY"
    dashscope:
      api_key_env: "DASHSCOPE_API_KEY"
      model: "qwen3-tts-instruct-flash-realtime"
      default_voice: "Cherry"
    minimax:
      api_key_env: "MINIMAX_API_KEY"
      model: "speech-2.8-turbo"

  # Per-persona voice mapping (provider-specific)
  # When switching TTS provider, update these values accordingly
  voice_map:
    ember: "gentle_female"
    iris: "Momo"
    kai: "calm_male"
    kelly: "calm_male"
    luna: "Maia"
    mia: "sweet_female"
    nova: "Cherry"
    rex: "calm_male"
    sora: "gentle_female"
    vivian: "Serena"

# ── Memory (双层记忆) ──
# SoulMem: 行为记忆 — 指导怎么做 (常驻，SQLite FTS5)
# EverMemOS: 陈述记忆 — 记住做了什么 (可选增强，HTTP)
memory:
  soulmem:
    db_path: ".data/memory.db"     # 常驻，总是启用

  evermemos:
    enabled: false                  # Set to true + base_url after starting EverMemOS
    base_url: ""                    # e.g. "http://localhost:1995/api/v1"
    api_key_env: "EVERMEMOS_API_KEY"

# ── Image (图片生成) ──
image:
  provider: "gemini"
  cache_dir: ".cache/image"
  providers:
    gemini:
      api_key_env: "GEMINI_API_KEY"
      model: "gemini-3.1-flash-image-preview"

# ── Speech (语音能力) ──
# 注: providers/config.py 同时支持 "tts" 和 "speech" key
# 当前保持 "tts" 不改，后续迁移到 "speech"

# ── 预留: 未来 API ──
# speech:
#   asr:
#     provider: "paraformer"
#   live:
#     provider: "realtime"