wangweiandCopilot
49ee50c104
Critical: the MCP SDK auto-enables DNS-rebinding protection when its host parameter is left at the 127.0.0.1 default, hard-coding a loopback-only Host allow-list. Every remote client (the only deployment this feature targets) was refused with HTTP 421 before auth or the tool ran. Now driven by a new MCP_ALLOWED_HOSTS setting, with '*' as an explicit, logged opt-out. Also bounds query/top_k to match AskRequest (top_k is amplified 4x downstream, so an unbounded value was a resource-exhaustion vector), decodes the Authorization header as latin-1 per the ASGI spec instead of raising a 500 on malformed bytes, and returns WWW-Authenticate on 401 per RFC 7235. Moves the psycopg2 import guard into backend/tests/conftest.py: duplicated across four test modules, it only worked because of alphabetical collection order, and any earlier-sorting package would have reintroduced a live connection attempt against the production database. Registers the mcp module in the authoritative backend architecture doc. 84 backend tests pass. Verified against a live server: allowed remote Host returns a valid initialize result, unknown Host returns 421, missing token returns 401 with WWW-Authenticate. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
179 lines
6.4 KiB
Bash
179 lines
6.4 KiB
Bash
# .env.example - 环境变量配置示例
|
||
# AI+合规智能中枢
|
||
|
||
# ===== 应用基础配置 =====
|
||
APP_NAME=AI+合规智能中枢
|
||
APP_VERSION=0.1.0
|
||
DEBUG=false
|
||
|
||
# ===== Milvus向量数据库配置 =====
|
||
MILVUS_HOST=6.86.80.8
|
||
MILVUS_PORT=19530
|
||
MILVUS_COLLECTION=regulations_dense_1024_v2
|
||
MILVUS_DB_NAME=default
|
||
MILVUS_INDEX_TYPE=IVF_FLAT
|
||
MILVUS_NLIST=128
|
||
MILVUS_NPROBE=16
|
||
|
||
# ===== 嵌入模型配置 =====
|
||
EMBEDDING_MODEL=text-embedding-v3
|
||
EMBEDDING_DIM=1024
|
||
EMBEDDING_API_KEY=your_embedding_api_key_here
|
||
EMBEDDING_BASE_URL=http://6.86.80.4:30080/v1
|
||
EMBEDDING_TIMEOUT_SECONDS=120
|
||
|
||
# ===== MinIO对象存储配置 =====
|
||
MINIO_ENDPOINT=6.86.80.8:9000
|
||
MINIO_ACCESS_KEY=minioadmin
|
||
MINIO_SECRET_KEY=minioadmin123
|
||
MINIO_BUCKET=compliance-docs
|
||
MINIO_SECURE=false
|
||
|
||
# ===== Redis配置 =====
|
||
REDIS_HOST=6.86.80.8
|
||
REDIS_PORT=6379
|
||
REDIS_PASSWORD=
|
||
REDIS_DB=0
|
||
|
||
# ===== PostgreSQL配置 =====
|
||
POSTGRES_HOST=6.86.80.8
|
||
POSTGRES_PORT=5432
|
||
POSTGRES_USER=compliance
|
||
POSTGRES_PASSWORD=compliance123
|
||
POSTGRES_DB=compliance_db
|
||
|
||
# ===== 文档处理配置 =====
|
||
CHUNK_SIZE=512
|
||
CHUNK_OVERLAP=50
|
||
MAX_FILE_SIZE_MB=100
|
||
DOCUMENT_METADATA_PATH=backend/data/documents.json
|
||
PARSER_BACKEND=aliyun
|
||
CHUNK_BACKEND=aliyun
|
||
# 文档元数据存储后端：json（默认，无需数据库）或 postgres（启用 PG 持久化）
|
||
# ⚠ 以下功能需要 postgres（设为 json 时功能静默降级或报 500）：
|
||
# - Direction B: 合规分析历史记录 (/compliance/history/*)
|
||
# - Direction B: DOCX 报告下载
|
||
# - Direction C: Finding Chat 消息持久化
|
||
DOCUMENT_REPOSITORY_BACKEND=json
|
||
# Set to true only when a Celery worker is running (./dev.sh start worker).
|
||
# Default false: document processing runs in FastAPI's threadpool (no external worker needed).
|
||
USE_CELERY_WORKER=false
|
||
|
||
# ===== 法规感知爬取配置 =====
|
||
PERCEPTION_CRAWL_TIMEOUT_SECONDS=120
|
||
PERCEPTION_MAX_EVENTS_PER_SOURCE=100
|
||
PERCEPTION_DIFF_SIMILARITY_THRESHOLD=0.85
|
||
|
||
# ===== 阿里云文档解析 =====
|
||
ALIBABA_ACCESS_KEY_ID=your_aliyun_access_key_id
|
||
ALIBABA_ACCESS_KEY_SECRET=your_aliyun_access_key_secret
|
||
ALIBABA_ENDPOINT=docmind-api.cn-hangzhou.aliyuncs.com
|
||
ALIYUN_PARSE_POLL_INTERVAL_SECONDS=5
|
||
ALIYUN_PARSE_TIMEOUT_SECONDS=900
|
||
ALIYUN_PARSE_LAYOUT_STEP_SIZE=50
|
||
ALIYUN_LLM_ENHANCEMENT=true
|
||
ALIYUN_ENHANCEMENT_MODE=VLM
|
||
DOCUMENT_PARSE_ARTIFACT_PREFIX=artifacts
|
||
PARSER_FAILURE_MODE=fail
|
||
|
||
# ===== API服务配置 =====
|
||
API_HOST=0.0.0.0
|
||
API_PORT=8000
|
||
|
||
# ===== LLM配置（必填）=====
|
||
# LLM提供商选择: qwen / deepseek / qwen_vl
|
||
LLM_PROVIDER=deepseek
|
||
LLM_MODEL=deepseek-v4-flash
|
||
LLM_MAX_TOKENS=4096
|
||
LLM_TEMPERATURE=0.7
|
||
|
||
# ===== 统一API代理配置 =====
|
||
# 使用new-api代理服务，支持多个LLM模型
|
||
# 获取API Key: 向管理员申请
|
||
QWEN_API_KEY=your_api_key_here
|
||
DEEPSEEK_API_KEY=your_api_key_here
|
||
QWEN_BASE_URL=http://6.86.80.4:30080/v1
|
||
DEEPSEEK_BASE_URL=http://6.86.80.4:30080/v1
|
||
|
||
# ===== 可用模型 =====
|
||
# Qwen系列: qwen3.5-plus, qwen3-plus, qwen-max, qwen-turbo, qwen-long
|
||
# Qwen VL系列: qwen3-vl-plus, qwen-vl-max
|
||
# DeepSeek系列: deepseek-v4-flash, deepseek-v3.2, deepseek-v3, deepseek-chat, deepseek-coder
|
||
QWEN_MODEL=qwen3.6-plus
|
||
QWEN_VL_MODEL=qwen3-vl-plus
|
||
DEEPSEEK_MODEL=deepseek-v4-flash
|
||
|
||
# ===== RAG配置 =====
|
||
RAG_TOP_K=10
|
||
RAG_RETRIEVAL_TOP_K=20
|
||
RAG_MAX_CONTEXT_TOKENS=4000
|
||
RAG_SUMMARY_MAX_TOKENS=1024
|
||
RAG_SKILLS_MAX_TOKENS=2048
|
||
|
||
# ── Reranker (Cross-Encoder) ──────────────────────────────────────────────────
|
||
# Set RERANKER_ENABLED=true and point to a TEI or Cohere-compatible rerank API.
|
||
# Recommended model: BAAI/bge-reranker-v2.5-gemma2-lightweight (lighter) or
|
||
# BAAI/bge-reranker-v2-m3 (heavier, higher quality).
|
||
# The endpoint must expose POST /rerank (TEI style) or POST /v1/rerank (Cohere style).
|
||
RERANKER_ENABLED=true
|
||
RERANKER_BASE_URL=http://6.86.80.4:30080/v1
|
||
RERANKER_MODEL=BAAI/bge-reranker-v2-m3
|
||
RERANKER_API_KEY=
|
||
RERANKER_TOP_K=5
|
||
|
||
# ===== 会话配置 =====
|
||
SESSION_MAX_SESSIONS=100
|
||
SESSION_TIMEOUT_MINUTES=30
|
||
# SESSION_BACKEND=redis 启用 Redis 持久化会话（需要 Redis 可用，推荐生产环境）
|
||
# SESSION_BACKEND=memory 使用内存会话（重启丢失，适合本地开发）
|
||
SESSION_BACKEND=memory
|
||
|
||
# ===== 认证配置 (Auth) =====
|
||
# 生产环境必须替换为强随机密钥:
|
||
# python -c "import secrets; print(secrets.token_hex(32))"
|
||
AUTH_SECRET_KEY=change-me-in-production-must-be-32-or-more-characters-long
|
||
AUTH_ALGORITHM=HS256
|
||
# Token 有效期（分钟），默认 8 小时
|
||
AUTH_TOKEN_EXPIRE_MINUTES=480
|
||
# 设为 false 可跳过认证（仅限本地开发调试，生产必须 true）
|
||
AUTH_ENABLED=true
|
||
|
||
|
||
# ===== HyDE 查询增强 =====
|
||
# HyDE (Hypothetical Document Embeddings): 在检索前让 LLM 生成一段"假设性回答"，
|
||
# 用该段落的 embedding 代替原始查询 embedding 进行向量检索。
|
||
# 无需新模型，复用现有 LLM 和 Embedding 服务。降低此功能可减少每次查询的 LLM 调用次数。
|
||
HYDE_ENABLED=true
|
||
HYDE_MAX_TOKENS=200
|
||
# ?????? LLM;???????????????
|
||
HYDE_LLM_PROVIDER=qwen
|
||
HYDE_LLM_MODEL=qwen3.5-flash
|
||
|
||
# ===== Agentic RAG 配置 (P0-1) =====
|
||
# 以下参数控制 /api/v1/agent/agentic/stream 多步推理管线
|
||
# 意图分类: simple_qa / compare / multi_hop / ambiguous
|
||
# compare 和 multi_hop 触发查询分解，最多 AGENTIC_MAX_SUB_QUERIES 个子查询
|
||
AGENTIC_MAX_SUB_QUERIES=4
|
||
# 引文锚定 fast-path 阈值: avg_score > 此值 且 chunks >= 3 时跳过 LLM grounding check
|
||
# 降低此值可让更多查询触发 LLM 二次验证（更准确，但延迟+成本增加）
|
||
AGENTIC_GROUNDING_THRESHOLD=0.65
|
||
# 各步骤 LLM 最大 token 数（越小越快，越大越准）
|
||
AGENTIC_INTENT_MAX_TOKENS=200
|
||
AGENTIC_PLAN_MAX_TOKENS=400
|
||
AGENTIC_GROUNDING_MAX_TOKENS=250
|
||
|
||
# ===== CORS =====
|
||
# 逗号分隔的允许跨域来源列表，生产环境绝不能使用 *
|
||
CORS_ALLOW_ORIGINS=http://localhost:5173
|
||
|
||
# ===== MCP (Model Context Protocol) =====
|
||
# MCP 端点（/mcp/）的 Host 头白名单，逗号分隔。MCP SDK 默认开启 DNS rebinding
|
||
# 防护，任何不在此列表中的 Host 都会被直接返回 HTTP 421，请求根本到不了鉴权和
|
||
# 工具逻辑。因此**远程部署必须把真实访问地址写进来**，否则所有外部 MCP 客户端
|
||
# （Claude Desktop / IDE 等）100% 连不上。
|
||
# 语法：`:*` 后缀表示匹配任意端口；填 `*` 表示彻底关闭该防护（不推荐）。
|
||
# 例如部署在 6.86.80.9:8000 时：
|
||
# MCP_ALLOWED_HOSTS=6.86.80.9:*,127.0.0.1:*,localhost:*
|
||
MCP_ALLOWED_HOSTS=127.0.0.1:*,localhost:*,[::1]:*
|
||
|