life-echo/api/app/core/config.py

"""
统一配置：所有环境变量通过此模块的 Settings 单点读取。
业务代码只允许 import settings，禁止散落 os.getenv() / load_dotenv()。

本地开发时由 api/development.sh 在启动前将 .env.development 同步为 .env（每次启动覆盖）。
Docker / 服务端由镜像与 compose 注入进程环境；此处仅固定读取工作目录下的 .env 作为默认值来源。
进程环境变量（容器 environment、export）覆盖 .env 同名项。
"""

import secrets

from pydantic import Field, field_validator
from pydantic_settings import BaseSettings, SettingsConfigDict


class Settings(BaseSettings):
    model_config = SettingsConfigDict(
        env_file=".env",
        env_file_encoding="utf-8",
        case_sensitive=False,
        extra="ignore",
    )

    # ── Database ──────────────────────────────────────────────
    database_url: str = "postgresql://postgres:postgres@localhost:5432/life_echo"
    # 启动时是否执行 Alembic（main.py lifespan）；测试或仅读场景可关
    alembic_run_on_startup: bool = True
    # True：迁移失败则进程退出（生产推荐）。False：仅打错误日志并继续（本地无 DB 时）
    alembic_startup_fail_fast: bool = False
    alembic_startup_max_retries: int = Field(default=3, ge=1, le=10)
    alembic_startup_retry_base_seconds: float = Field(default=1.0, ge=0.1, le=60.0)

    # ── Redis ─────────────────────────────────────────────────
    redis_url: str = "redis://localhost:6379/0"
    redis_session_ttl: int = 86400

    # ── Auth / JWT ────────────────────────────────────────────
    secret_key: str = Field(default_factory=lambda: secrets.token_urlsafe(32))
    algorithm: str = "HS256"
    access_token_expire_minutes: int = 120
    refresh_token_expire_days: int = 30

    # ── LLM / DeepSeek ───────────────────────────────────────
    deepseek_api_key: str = ""
    deepseek_base_url: str = "https://api.deepseek.com"
    deepseek_model: str = "deepseek-chat"
    llm_api_key: str = ""
    llm_base_url: str = ""
    llm_model: str = ""
    llm_temperature: float = 0.7
    # 空字符串：快档位与默认模型相同；分类/抽取/记忆富化等可单独指定较轻模型
    llm_fast_model: str = ""

    # ── Memory 向量（智谱 BigModel 国内 embedding-3；与 LLM/DeepSeek 密钥分离）──
    zhipu_api_key: str = ""
    embedding_base_url: str = "https://open.bigmodel.cn/api/paas/v4"
    embedding_model: str = "embedding-3"

    # ── Chat 访谈（token 上限 + 代码截断，见 reply_limits）──
    chat_interview_max_tokens: int = 380
    chat_interview_max_segments: int = 2
    chat_interview_max_chars_per_segment: int = 260
    # 访谈：用户本轮极短输入时的更紧上限（见 interview_reply_length）
    chat_interview_brief_max_tokens: int = Field(default=260, ge=64, le=2048)
    chat_interview_brief_max_chars_per_segment: int = Field(default=200, ge=32, le=2000)
    # 访谈：有新细节/情绪/长段时的展开上限
    chat_interview_expanded_max_tokens: int = Field(default=520, ge=64, le=4096)
    chat_interview_expanded_max_chars_per_segment: int = Field(
        default=380, ge=32, le=4000
    )
    # 干部/军队推断命中时，standard 档在分桶基础上小幅放宽（brief/expanded 不变）
    chat_interview_cadre_military_standard_extra_tokens: int = Field(
        default=40, ge=0, le=512
    )
    chat_interview_cadre_military_standard_extra_chars: int = Field(
        default=40, ge=0, le=2000
    )
    chat_opening_max_tokens: int = 256
    chat_profile_followup_max_tokens: int = 280
    # Redis 全量历史仅用于 turn 计数；注入 LLM 时截取最近若干轮与字符预算
    chat_history_max_pairs: int = Field(default=15, ge=1, le=500)
    chat_history_max_chars: int = Field(default=6000, ge=256, le=500_000)
    chat_era_context_enabled: bool = True
    # 访谈：每轮用 LLM 判定用户主人生阶段并更新 MemoirState.current_stage；False 时仅用关键词
    chat_stage_detection_enabled: bool = True
    chat_stage_detection_max_tokens: int = 128
    # 访谈性格：default | warm_listener | curious_guide（未知值按 default）
    chat_interview_persona: str = "default"
    # 访谈：按用户本轮话检索记忆并注入 prompt（关则不调 MemoryService.retrieve）
    chat_memory_retrieval_enabled: bool = True
    chat_memory_top_k: int = Field(default=8, ge=1, le=30)
    chat_memory_evidence_max_chars: int = Field(default=4096, ge=256, le=50_000)

    # ── Memoir 叙事忠实度检查（FidelityCheckAgent）────────────────
    memoir_fidelity_check_enabled: bool = True
    memoir_fidelity_check_max_tokens: int = 512
    # 口述归一（进入叙事 / 忠实度前；segment 原文不落库）：off | rules | llm
    memoir_oral_normalize_enabled: bool = True
    memoir_oral_normalize_mode: str = "rules"
    memoir_oral_normalize_llm_max_tokens: int = Field(default=512, ge=64, le=4096)
    memoir_oral_normalize_llm_max_input_chars: int = Field(
        default=8000, ge=64, le=50_000
    )
    # 聊天：模型消费净稿（不改变 segment 落库原文）；与 memoir 规则层共用，配置独立
    chat_input_normalize_enabled: bool = True
    chat_input_normalize_mode: str = "rules"  # off | rules | llm
    chat_input_normalize_llm_max_tokens: int = Field(default=512, ge=64, le=4096)
    chat_input_normalize_llm_max_input_chars: int = Field(
        default=8000, ge=64, le=50_000
    )

    # ── ASR ───────────────────────────────────────────────────
    asr_provider: str = "whisper"
    asr_model_size: str = "small"
    asr_device: str = "auto"
    asr_compute_type: str = "auto"
    asr_model_cache_dir: str = ""

    # ── Tencent SMS ──────────────────────────────────────────
    tencent_sms_secret_id: str = ""
    tencent_sms_secret_key: str = ""
    tencent_sms_sdk_app_id: str = ""
    tencent_sms_sign_name: str = ""
    tencent_sms_template_id: str = ""
    tencent_sms_template_param_count: int = 2

    # ── Tencent ASR / TTS（共用 Secret；与短信、COS 密钥独立）────────────────
    tencent_secret_id: str = ""
    tencent_secret_key: str = ""

    # ── TTS (openai | tencent)，与 ASR 独立：仅控制回复侧语音合成 ──
    enable_tts: bool = True
    tts_provider: str = "tencent"
    openai_api_key: str = ""
    tts_voice_type: int = 502001  # Tencent 音色 ID，见 https://cloud.tencent.com/document/product/1073/92668
    tts_codec: str = "mp3"

    # ── WeChat Pay ───────────────────────────────────────────
    wechat_pay_app_id: str = ""
    wechat_pay_mch_id: str = ""
    wechat_pay_api_v3_key: str = ""
    wechat_pay_private_key_path: str = "certs/apiclient_key.pem"
    wechat_pay_private_key: str = ""  # PEM 内容，与 private_key_path 二选一
    wechat_pay_cert_serial_no: str = ""
    wechat_pay_notify_url: str = ""
    wechat_pay_platform_public_key: str = ""
    wechat_pay_platform_public_key_path: str = ""
    wechat_pay_platform_public_key_id: str = ""

    # ── Alipay ───────────────────────────────────────────────
    alipay_app_id: str = ""
    alipay_private_key: str = ""
    alipay_public_key: str = ""
    alipay_notify_url: str = ""
    alipay_sign_type: str = "RSA2"
    alipay_under_development: str = "true"  # "1"/"true"/"yes" 视为开发中不可用

    # ── Logging ──────────────────────────────────────────────
    # 环境变量 LOG_LEVEL；控制 loguru sink 最低级别（TRACE/DEBUG/INFO/…）
    log_level: str = "INFO"
    # LOG_AGENT_VERBOSE：为 True 时额外输出 Agent 单行 INFO 摘要（耗时、规模），无需全局 DEBUG
    log_agent_verbose: bool = False
    # AGENT_LOG_MAX_CHARS：DEBUG 下记录 prompt/响应预览时的最大字符数
    agent_log_max_chars: int = Field(default=4096, ge=256, le=100_000)
    # 第三方 stdlib logging（空=自动：LOG_LEVEL 为 DEBUG/TRACE 时 Celery→INFO、httpx/httpcore→WARNING）
    celery_log_level: str = ""
    httpx_log_level: str = ""

    @field_validator("log_agent_verbose", mode="before")
    @classmethod
    def _coerce_log_agent_verbose(cls, v: object) -> bool:
        if isinstance(v, bool):
            return v
        if v is None:
            return False
        return str(v).strip().lower() in ("1", "true", "yes", "on")

    # ── Misc ─────────────────────────────────────────────────
    enable_test_subscription: int = 0
    enable_test_plan: str = ""  # "1" / "true" / "yes" 为 True
    enable_docs: bool = True

    # ── Memoir Image ─────────────────────────────────────────
    memoir_image_enabled: bool = False
    # True：图片 LLM prompt 失败时不使用英语降级模板（需产品与任务失败流确认后开启）
    image_prompt_fallback_disabled: bool = False
    memoir_image_poll_interval: int = 3
    memoir_image_max_attempts: int = 20
    memoir_image_provider: str = "liblib"
    memoir_image_style_default: str = "watercolor"
    memoir_image_size_default: str = "1280x720"
    memoir_image_download_hosts: str = ""
    # Story 正文至少多少字才创建主图 intent / 调图（0 表示不限制）
    story_image_min_body_chars: int = 400
    # generate_story_image 入队去重（Redis SET NX，秒）
    story_image_enqueue_dedup_ttl: int = Field(default=300, ge=30, le=86400)
    # 章节物化异步任务延迟入队（秒），削峰
    recompose_chapter_delay_seconds: int = Field(default=8, ge=0, le=600)
    # 与 memoir pipeline 一致的章节互斥锁 TTL（秒）
    chapter_pipeline_lock_ttl_seconds: int = Field(default=120, ge=10, le=3600)
    # Append 硬上限：canonical 字符数、版本数（超限强制 new_story）
    story_append_max_canonical_chars: int = Field(default=12000, ge=1000, le=500_000)
    story_append_max_versions: int = Field(default=20, ge=1, le=500)
    # Evidence 检索 top_k：大批次 unit 时降低检索量
    evidence_top_k_default: int = Field(default=10, ge=1, le=50)
    evidence_top_k_large_batch: int = Field(default=5, ge=1, le=50)
    evidence_large_batch_threshold: int = Field(default=3, ge=1, le=100)
    # Story/Chapter 标题在正文达到此字数后才由 LLM 生成；之前用占位符
    story_title_min_body_chars: int = Field(default=60, ge=0, le=10_000)
    # 回忆录 Celery：累计 strip 后口述字数未达此值则暂缓提交（0=关闭，仅防抖后提交）
    memoir_segment_batch_min_chars: int = Field(default=50, ge=0, le=50_000)
    # 本批首条 segment 入队起最长等待（秒），超时则提交（即使字数不足）
    memoir_segment_batch_max_wait_seconds: float = Field(
        default=60.0, ge=0.0, le=3600.0
    )

    # ── Memory 检索与富化 ─────────────────────────────────────
    # True：query 为空时仍返回 rolling 摘要 + 最近事实/时间线（无 chunk FTS）
    memory_evidence_empty_query_include_rolling: bool = False
    # False：跳过 ingest 后 LLM 富化（摘要/事实/时间线）
    memory_enrichment_enabled: bool = True
    memory_enrichment_max_chars: int = Field(default=12000, ge=1000, le=100_000)

    # ── Memory compaction（近重复 chunk 软排除；事件触发 + Redis 防抖 + 用户锁）──
    memory_compaction_enabled: bool = False
    memory_compaction_debounce_seconds: int = Field(default=105, ge=10, le=3600)
    memory_compaction_lock_ttl_seconds: int = Field(default=600, ge=60, le=7200)
    memory_compaction_chunk_similarity_threshold: float = Field(
        default=0.92, ge=0.5, le=0.999
    )
    memory_compaction_min_layers_for_exclude: int = Field(default=2, ge=1, le=3)
    memory_compaction_max_chunks_per_run: int = Field(default=200, ge=1, le=10_000)
    memory_compaction_max_excludes_per_run: int = Field(default=50, ge=1, le=1000)
    memory_compaction_max_neighbors_per_chunk: int = Field(default=25, ge=5, le=100)
    memory_compaction_text_jaccard_min: float = Field(default=0.55, ge=0.0, le=1.0)
    memory_compaction_metadata_event_year_window: int = Field(default=1, ge=0, le=50)

    # ── Liblib ───────────────────────────────────────────────
    liblib_access_key: str = ""
    liblib_secret_key: str = ""
    liblib_base_url: str = "https://openapi.liblibai.cloud"
    liblib_template_uuid: str = ""

    # ── Tencent COS ──────────────────────────────────────────
    tencent_cos_secret_id: str = ""
    tencent_cos_secret_key: str = ""
    tencent_cos_region: str = "ap-shanghai"
    tencent_cos_bucket: str = ""
    tencent_cos_base_url: str = ""
    tencent_cos_token: str = ""


settings = Settings()