""" 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 ) # 回忆录叙事 Phase 2（ Celery）触发：单条口述达到该 strip 字数则立即跑叙事 memoir_narrative_immediate_char_threshold: int = Field(default=50, ge=0, le=50_000) # 同一 topic_category 下未叙事段数达到该值则触发 Phase 2 memoir_narrative_batch_min_segments: int = Field(default=3, ge=1, le=500) # 同上，累计 user_input_text 字符数（strip 后由 Segment 列 length 近似） memoir_narrative_batch_min_chars: int = Field(default=80, ge=0, le=500_000) # Phase 1 完成后未触发 Phase 2 时，延迟任务兜底（秒）；新 Phase 1 会 revoke 旧定时 memoir_narrative_batch_max_wait_seconds: float = Field( default=120.0, ge=1.0, le=3600.0 ) # False：Celery/批处理更新 slot 时不改写 MemoirState.current_stage（访谈路径仍可由 switch_stage 推进） # True：仅当 chat_bucket( proposed ) == chat_bucket( existing ) 时允许批处理对齐 current_stage memoir_extraction_updates_current_stage: bool = False # ── 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()