125 lines
5.1 KiB
Python
125 lines
5.1 KiB
Python
import os
|
|
from dataclasses import dataclass, field
|
|
from pathlib import Path
|
|
|
|
from dotenv import load_dotenv
|
|
|
|
load_dotenv(Path(__file__).resolve().parent.parent / ".env")
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class Settings:
|
|
# LLM 配置
|
|
llm_api_key: str = field(
|
|
default_factory=lambda: os.getenv("LLM_API_KEY", ""),
|
|
repr=False,
|
|
)
|
|
llm_model: str = field(default_factory=lambda: os.getenv("LLM_MODEL", ""))
|
|
llm_base_url: str | None = field(
|
|
default_factory=lambda: os.getenv("LLM_BASE_URL") or None
|
|
)
|
|
llm_timeout: float = 60.0
|
|
llm_max_retries: int = 2
|
|
|
|
# Jev
|
|
typesafe_api_key: str = field(
|
|
default_factory=lambda: os.getenv("TYPESAFE_API_KEY", ""), repr=False
|
|
)
|
|
typesafe_model: str = field(
|
|
default_factory=lambda: os.getenv("TYPESAFE_DEFAULT_MODEL") or "jev-latest"
|
|
)
|
|
typesafe_base_url: str = field(
|
|
default_factory=lambda: (
|
|
os.getenv("TYPESAFE_BASE_URL") or "https://api.typesafe.ai"
|
|
)
|
|
)
|
|
typesafe_timeout: float = field(
|
|
default_factory=lambda: float(os.getenv("TYPESAFE_TIMEOUT", "30"))
|
|
)
|
|
typesafe_max_retries: int = field(
|
|
default_factory=lambda: int(os.getenv("TYPESAFE_MAX_RETRIES", "2"))
|
|
)
|
|
|
|
# 意图门控：Jev 选中 qa 且概率达到阈值才走问答，其余都交给 screen agent；阈值按真实语料校准
|
|
intent_qa_min_probability: float = field(
|
|
default_factory=lambda: float(os.getenv("INTENT_QA_MIN_PROBABILITY", "0.5"))
|
|
)
|
|
|
|
# screen agent 单次运行的模型调用上限，包括读取 skill、查询物料、提交修正和最终回复
|
|
max_agent_steps: int = field(
|
|
default_factory=lambda: int(os.getenv("MAX_AGENT_STEPS", "12"))
|
|
)
|
|
# 单次 MCP 工具调用的超时（秒），服务列表见 mcp_servers.json
|
|
mcp_tool_timeout: float = field(
|
|
default_factory=lambda: float(os.getenv("MCP_TOOL_TIMEOUT", "5"))
|
|
)
|
|
|
|
# 没有任何对话的空会话保留时长（秒），超时后视为不存在
|
|
conversation_empty_ttl: int = field(
|
|
default_factory=lambda: int(os.getenv("CONVERSATION_EMPTY_TTL", "7200"))
|
|
)
|
|
# 每次请求从会话记录读取的最近轮数上限，实际带入多少由 history_max_tokens 决定
|
|
conversation_history_turns: int = field(
|
|
default_factory=lambda: int(os.getenv("CONVERSATION_HISTORY_TURNS", "20"))
|
|
)
|
|
# 每次请求带入的会话历史 token 预算（近似计数），从最近一轮往前保留完整的轮次
|
|
history_max_tokens: int = field(
|
|
default_factory=lambda: int(os.getenv("HISTORY_MAX_TOKENS", "4096"))
|
|
)
|
|
|
|
# MongoDB 连接串，未配置时会话记录和检查点回退为进程内存实现
|
|
mongodb_uri: str | None = field(
|
|
default_factory=lambda: os.getenv("MONGODB_URI") or None, repr=False
|
|
)
|
|
mongodb_db: str = field(
|
|
default_factory=lambda: os.getenv("MONGODB_DB") or "screen_agent"
|
|
)
|
|
# 检查点保留时长（秒），超时由 MongoDB TTL 索引清理，0 表示永久保留
|
|
mongodb_checkpoint_ttl: int = field(
|
|
default_factory=lambda: int(os.getenv("MONGODB_CHECKPOINT_TTL", "604800"))
|
|
)
|
|
|
|
# 用户中心：请求携带的 token 交给它校验，换回 user_id
|
|
user_center_url: str = field(
|
|
default_factory=lambda: (
|
|
os.getenv("USER_CENTER_URL") or "http://127.0.0.1:8100"
|
|
).rstrip("/")
|
|
)
|
|
user_center_timeout_seconds: float = field(
|
|
default_factory=lambda: float(os.getenv("USER_CENTER_TIMEOUT_SECONDS", "5"))
|
|
)
|
|
# 关闭后不校验 token，所有请求视为同一个本地调试用户，仅用于开发
|
|
auth_enabled: bool = field(
|
|
default_factory=lambda: (
|
|
os.getenv("AUTH_ENABLED", "true").strip().lower()
|
|
not in ("false", "0", "no", "off")
|
|
)
|
|
)
|
|
# token 校验结果缓存时长（秒），登出或禁用后最多延迟这么久失效，0 表示不缓存
|
|
auth_cache_ttl_seconds: int = field(
|
|
default_factory=lambda: int(os.getenv("AUTH_CACHE_TTL_SECONDS", "30"))
|
|
)
|
|
|
|
def __post_init__(self):
|
|
if self.user_center_timeout_seconds <= 0:
|
|
raise ValueError("USER_CENTER_TIMEOUT_SECONDS 必须是正数")
|
|
if self.auth_cache_ttl_seconds < 0:
|
|
raise ValueError("AUTH_CACHE_TTL_SECONDS 必须是非负整数")
|
|
if self.mongodb_checkpoint_ttl < 0:
|
|
raise ValueError("MONGODB_CHECKPOINT_TTL 必须是非负整数")
|
|
if self.conversation_empty_ttl <= 0:
|
|
raise ValueError("CONVERSATION_EMPTY_TTL 必须是正整数")
|
|
if self.conversation_history_turns < 0:
|
|
raise ValueError("CONVERSATION_HISTORY_TURNS 必须是非负整数")
|
|
if self.history_max_tokens < 1:
|
|
raise ValueError("HISTORY_MAX_TOKENS 必须是正整数")
|
|
if self.max_agent_steps < 1:
|
|
raise ValueError("MAX_AGENT_STEPS 必须是正整数")
|
|
if self.mcp_tool_timeout <= 0:
|
|
raise ValueError("MCP_TOOL_TIMEOUT 必须是正数")
|
|
if not 0 <= self.intent_qa_min_probability <= 1:
|
|
raise ValueError("INTENT_QA_MIN_PROBABILITY 必须在 0 到 1 之间")
|
|
|
|
|
|
settings = Settings()
|