2fded6fbd9
- Drop interview_reply_length and utterance_substance; always run stage LLM and memory retrieval when enabled; trim Settings fields and .env.example. - Replace guided/opening prompts with compact fact blocks plus unified behavior guidance; slim background_voice and persona to tone hints. - InterviewAgent uses fixed chat_interview max_tokens/chars/segments. Also includes stacked work: profile followup/extract path, evaluation rubric and judge schema updates, transcript SPLIT handling in execution service, user export markdown split tests, and golden case fixture.
336 lines
19 KiB
Python
336 lines
19 KiB
Python
"""
|
||
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
|
||
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
|
||
)
|
||
# True 且 mode=llm：仅语音/ASR 段走 LLM 纠错；键盘输入仅规则归一（省每轮 LLM）
|
||
chat_input_normalize_llm_voice_only: bool = True
|
||
# 资料收集：超过该对话轮次（Redis 全量轮次计数）仍有缺失字段时，强制进入访谈，避免长期问卷感
|
||
chat_profile_max_turns: int = Field(default=8, ge=1, le=500)
|
||
|
||
# Memoir Phase1：多 segment 一批一次 LLM 完成抽取+章节分类（失败回退逐段）；单段且关时仍逐段
|
||
memoir_phase1_batch_llm_enabled: bool = False
|
||
memoir_phase1_batch_llm_max_tokens: int = Field(default=4096, ge=512, le=32_768)
|
||
# Memoir agents：`invoke_json_object` / `llm_json_call` 的 max_tokens（原硬编码迁至配置）
|
||
memoir_extraction_max_tokens: int = Field(default=1024, ge=64, le=8192)
|
||
memoir_classification_max_tokens: int = Field(default=256, ge=32, le=4096)
|
||
memoir_narrative_max_tokens: int = Field(default=4096, ge=256, le=32_768)
|
||
memoir_narrative_merge_max_tokens: int = Field(default=8192, ge=256, le=64_000)
|
||
memoir_title_max_tokens: int = Field(default=256, ge=32, le=4096)
|
||
memoir_story_route_max_tokens: int = Field(default=1024, ge=64, le=8192)
|
||
memoir_story_batch_plan_max_tokens: int = Field(default=4096, ge=256, le=32_768)
|
||
# 资料抽取（ProfileAgent JSON 模式）
|
||
chat_profile_extract_max_tokens: int = Field(default=512, ge=64, le=4096)
|
||
|
||
# ── 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)
|
||
# AGENT_LOG_OMIT_SYSTEM_MESSAGE_BODY：DEBUG 下访谈/资料聊天日志省略 System 正文（仅 len+sha12）
|
||
agent_log_omit_system_message_body: bool = True
|
||
# AGENT_LOG_JSON_PROMPT_PREFIX_CHARS：DEBUG 下 *.prompt 总长超过下项时再跳过前 N 字符后预览（0=不跳过）
|
||
agent_log_json_prompt_prefix_chars: int = Field(default=0, ge=0, le=500_000)
|
||
# AGENT_LOG_JSON_PROMPT_PREFIX_ONLY_IF_LEN_GT：触发“跳过前缀”的最小 prompt 长度
|
||
agent_log_json_prompt_prefix_only_if_len_gt: int = Field(
|
||
default=4000, ge=0, le=2_000_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")
|
||
|
||
@field_validator("agent_log_omit_system_message_body", mode="before")
|
||
@classmethod
|
||
def _coerce_agent_log_omit_system_message_body(cls, v: object) -> bool:
|
||
if isinstance(v, bool):
|
||
return v
|
||
if v is None:
|
||
return True
|
||
s = str(v).strip().lower()
|
||
if s in ("0", "false", "no", "off"):
|
||
return False
|
||
return True
|
||
|
||
# ── 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）；应覆盖 Phase2 / recompose 的 P95 时长
|
||
chapter_pipeline_lock_ttl_seconds: int = Field(default=360, 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)
|
||
# StoryRouteAgent：候选 JSON 预算（保守默认，可调大）
|
||
story_route_candidate_body_max_chars: int = Field(default=1600, ge=200, le=8000)
|
||
story_route_candidate_total_max_chars: int = Field(
|
||
default=16_000, ge=2000, le=100_000
|
||
)
|
||
story_route_long_body_head_chars: int = Field(default=700, ge=100, le=4000)
|
||
story_route_long_body_tail_chars: int = Field(default=700, ge=100, le=4000)
|
||
story_route_summary_min_chars: int = Field(default=30, ge=0, le=500)
|
||
story_route_index_preview_chars: int = Field(default=80, ge=20, 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
|
||
# True：FidelityCheckAgent JSON/LLM 解析失败时放行（仅建议 append 场景配合 existing 兜底）
|
||
memoir_fidelity_fail_open_on_parse_error: bool = False
|
||
# 正文与 evidence 文本的最长公共子串达到该长度且 oral/旧正文未覆盖时，回退为安全正文
|
||
memoir_narrative_evidence_overlap_min_chars: int = Field(default=14, ge=8, le=256)
|
||
# True：启用短「场合锚点」词检测（聚餐/那晚等），须同时在摘录中出现且口述未覆盖才回退
|
||
memoir_evidence_scene_anchor_check_enabled: bool = True
|
||
# True：标题生成时 slots 仅保留在 oral 或正文摘录中出现的条目（减少档案串台）
|
||
memoir_title_slots_require_body_or_oral_match: bool = True
|
||
# True：标题中出现高置信「履历链」短语则须在 hay（正文+口述+已传 slots）中有逐字依据，否则降级占位
|
||
memoir_title_hay_grounding_strict_phrases_enabled: bool = True
|
||
# True：章节物化拿不到 pipeline 锁时 Celery retry（避免长期跳过导致 dirty 不收敛）
|
||
memoir_recompose_retry_on_lock_contention: bool = True
|
||
# Phase2 立即派发使用固定 task_id，减少同类目重复入队（超时任务仍用独立 id）
|
||
memoir_phase2_singleflight_immediate: bool = True
|
||
|
||
# ── 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 向量检索）
|
||
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)
|
||
# True：事实 ILIKE 未命中时退回「最近 confirmed 事实」（易引入无关/矛盾事实；默认关）
|
||
memory_fact_search_use_recent_fallback: bool = False
|
||
|
||
# ── Memory compaction（近重复 chunk 软排除；事件触发 + Redis 防抖 + 用户锁；需 worker + Beat 跑 sweep）──
|
||
memory_compaction_enabled: bool = True
|
||
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)
|
||
# Beat sweep：扫描最近 N 小时内有新 chunk 的用户并调度 compaction
|
||
memory_compaction_sweep_recent_hours: int = Field(default=24, ge=1, le=168)
|
||
|
||
# ── 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 = ""
|
||
|
||
# ── Internal regression evaluation lab（独立入口，不挂在消费者 API）────
|
||
internal_eval_api_key: str = ""
|
||
internal_eval_enable_docs: bool = False
|
||
# 逗号分隔；空则内部 API 不额外限制 Origin（仍可依赖 internal_eval_api_key）
|
||
internal_eval_cors_origins: str = ""
|
||
# GLM / 智谱：评审模型（OpenAI 兼容 Chat Completions，与 langchain-openai 一致）
|
||
eval_judge_api_key: str = ""
|
||
eval_judge_base_url: str = "https://open.bigmodel.cn/api/paas/v4"
|
||
eval_judge_model: str = "glm-4-flash"
|
||
eval_judge_temperature: float = 0.3
|
||
# 候选对话回放：与生产访谈类似的温度
|
||
eval_candidate_temperature: float = 0.7
|
||
# 门禁：受保护 session 合成份数下跌超过该阈值视为回归（0–100 分制）
|
||
eval_gate_protected_regression_threshold: float = Field(
|
||
default=2.0, ge=0.0, le=100.0
|
||
)
|
||
# 执行 LLM 判分与回放（Celery 未跑时可关，仅跑结构/导入）
|
||
eval_execution_enabled: bool = True
|
||
|
||
|
||
settings = Settings()
|