feat: configure RAG system with ModelScope Embedding + Jina Rerank

- Switch API config to environment variables (EMBED_API_KEY, RERANK_API_KEY)
- Use ModelScope Qwen3-Embedding-8B for vector embedding
- Use Jina jina-reranker-v3 for reranking
- Fix text/plain response handling for ModelScope API
- Remove unused LLM config
- Document RAG system in README

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

.claude/scripts/data_modules/api_client.py

@@ -88,7 +88,8 @@ class EmbeddingAPIClient:
         if self.config.embed_api_type == "openai":
             return {
                 "input": texts,
-                "model": self.config.embed_model
+                "model": self.config.embed_model,
+                "encoding_format": "float"
             }
         else:
             # Modal 格式
@@ -135,7 +136,9 @@ class EmbeddingAPIClient:
                     timeout=aiohttp.ClientTimeout(total=timeout)
                 ) as resp:
                     if resp.status == 200:
-                        data = await resp.json()
+                        text = await resp.text()
+                        import json as json_module
+                        data = json_module.loads(text)
                         embeddings = self._parse_response(data)
 
                         if embeddings:
@@ -144,7 +147,8 @@ class EmbeddingAPIClient:
                             return embeddings
 
                     self.stats.errors += 1
-                    print(f"[ERR] Embed {resp.status}: {await resp.text()[:200]}")
+                    err_text = await resp.text()
+                    print(f"[ERR] Embed {resp.status}: {err_text[:200]}")
                     return None
 
             except Exception as e:
@@ -303,7 +307,8 @@ class RerankAPIClient:
                         return self._parse_response(data)
                     else:
                         self.stats.errors += 1
-                        print(f"[ERR] Rerank {resp.status}: {await resp.text()[:200]}")
+                        err_text = await resp.text()
+                        print(f"[ERR] Rerank {resp.status}: {err_text[:200]}")
                         return None
 
             except Exception as e:

.claude/scripts/data_modules/config.py

@@ -2,6 +2,10 @@
 # -*- coding: utf-8 -*-
 """
 Data Modules - 配置文件
+
+API 配置通过环境变量读取：
+- EMBED_BASE_URL, EMBED_MODEL, EMBED_API_KEY
+- RERANK_BASE_URL, RERANK_MODEL, RERANK_API_KEY
 """
 
 import os
@@ -45,50 +49,36 @@ class DataModulesConfig:
     def outline_dir(self) -> Path:
         return self.project_root / "大纲"
 
-    # ================= Modal API Endpoints =================
-    # 注意：以下为默认 Modal 端点，可通过环境变量或显式传参覆盖
-    llm_base_url: str = "https://lingfengqaq--qwen3-30b-vllm-serve.modal.run/v1"
-    llm_model: str = "Qwen/Qwen3-30B-A3B-Instruct-2507"
 
     # ================= Embedding API 配置 =================
-    # api_type: "openai" (通用 OpenAI 兼容接口) | "modal" (Modal 自定义接口)
     embed_api_type: str = "openai"
-    embed_base_url: str = "https://lingfengqaq--qwen-embedding-server-qwenembedding-embeddings.modal.run"
-    embed_model: str = "qwen-embedding"
-    embed_api_key: str = ""  # OpenAI 兼容接口需要 API Key
+    embed_base_url: str = field(default_factory=lambda: os.getenv("EMBED_BASE_URL", "https://api-inference.modelscope.cn/v1"))
+    embed_model: str = field(default_factory=lambda: os.getenv("EMBED_MODEL", "Qwen/Qwen3-Embedding-8B"))
+    embed_api_key: str = field(default_factory=lambda: os.getenv("EMBED_API_KEY", ""))
 
-    # 保留旧字段兼容
     @property
     def embed_url(self) -> str:
-        """兼容旧代码：返回 embed_base_url"""
         return self.embed_base_url
 
     # ================= Rerank API 配置 =================
-    # api_type: "openai" (如 Jina/Cohere 兼容接口) | "modal" (Modal 自定义接口)
-    rerank_api_type: str = "modal"
-    rerank_base_url: str = "https://lingfengqaq--qwen-reranker-server-qwenreranker-rerank.modal.run"
-    rerank_model: str = "qwen-reranker"
-    rerank_api_key: str = ""  # Jina/Cohere 等需要 API Key
+    rerank_api_type: str = "openai"
+    rerank_base_url: str = field(default_factory=lambda: os.getenv("RERANK_BASE_URL", "https://api.jina.ai/v1"))
+    rerank_model: str = field(default_factory=lambda: os.getenv("RERANK_MODEL", "jina-reranker-v3"))
+    rerank_api_key: str = field(default_factory=lambda: os.getenv("RERANK_API_KEY", ""))
 
-    # 保留旧字段兼容
     @property
     def rerank_url(self) -> str:
-        """兼容旧代码：返回 rerank_base_url"""
         return self.rerank_base_url
 
     # ================= 并发配置 =================
-    llm_concurrency: int = 32
     embed_concurrency: int = 64
     rerank_concurrency: int = 32
     embed_batch_size: int = 64
 
     # ================= 超时配置 =================
-    cold_start_timeout: int = 300  # 5 分钟
-    normal_timeout: int = 180      # 3 分钟
+    cold_start_timeout: int = 300
+    normal_timeout: int = 180
 
-    # ================= LLM 生成配置 =================
-    llm_temperature: float = 0.1
-    llm_max_tokens: int = 4096
 
     # ================= 检索配置 =================
     vector_top_k: int = 30
@@ -96,9 +86,6 @@ class DataModulesConfig:
     rerank_top_n: int = 10
     rrf_k: int = 60
 
-    # 向量检索性能开关
-    # - 向量数量较少时（<= full_scan_max_vectors）可全表扫描，召回更稳
-    # - 规模变大后默认走预筛选（BM25 + 最近片段），避免 O(n) 扫描拖慢 Context Agent
     vector_full_scan_max_vectors: int = 500
     vector_prefilter_bm25_candidates: int = 200
     vector_prefilter_recent_candidates: int = 200
@@ -108,18 +95,15 @@ class DataModulesConfig:
     extraction_confidence_medium: float = 0.5
 
     # ================= 列表截断限制 =================
-    # state.json 列表最大保留条数
     max_disambiguation_warnings: int = 500
     max_disambiguation_pending: int = 1000
     max_state_changes: int = 2000
 
-    # Context Pack 输出切片
     context_recent_summaries_window: int = 5
     context_alerts_slice: int = 10
     context_max_appearing_characters: int = 10
     context_max_urgent_foreshadowing: int = 5
 
-    # 导出上下文时的列表截断
     export_recent_changes_slice: int = 20
     export_disambiguation_slice: int = 20
 
@@ -130,32 +114,29 @@ class DataModulesConfig:
     query_recent_appearances_limit: int = 20
 
     # ================= 伏笔紧急度 =================
-    # 紧急度阈值（基于 章节差 / 目标差 × 权重）
-    foreshadowing_urgency_pending_high: int = 100  # 超过 100 章未回收
-    foreshadowing_urgency_pending_medium: int = 50  # 超过 50 章
-    foreshadowing_urgency_target_proximity: int = 5  # 距目标章节 5 章内
+    foreshadowing_urgency_pending_high: int = 100
+    foreshadowing_urgency_pending_medium: int = 50
+    foreshadowing_urgency_target_proximity: int = 5
     foreshadowing_urgency_score_high: int = 100
     foreshadowing_urgency_score_medium: int = 60
     foreshadowing_urgency_score_target: int = 80
     foreshadowing_urgency_score_low: int = 20
-    foreshadowing_urgency_threshold_show: int = 60  # >= 此值才显示
+    foreshadowing_urgency_threshold_show: int = 60
 
-    # 层级权重
     foreshadowing_tier_weight_core: float = 3.0
     foreshadowing_tier_weight_sub: float = 2.0
     foreshadowing_tier_weight_decor: float = 1.0
 
     # ================= 角色活跃度 =================
-    character_absence_warning: int = 30  # 轻度掉线阈值
-    character_absence_critical: int = 100  # 严重掉线阈值
-    character_candidates_limit: int = 800  # 扫描时候选角色上限
+    character_absence_warning: int = 30
+    character_absence_critical: int = 100
+    character_candidates_limit: int = 800
 
     # ================= Strand Weave 节奏 =================
-    strand_quest_max_consecutive: int = 5  # Quest 线最大连续章数
-    strand_fire_max_gap: int = 10  # Fire 线最大缺失章数
-    strand_constellation_max_gap: int = 15  # Constellation 线最大缺失章数
+    strand_quest_max_consecutive: int = 5
+    strand_fire_max_gap: int = 10
+    strand_constellation_max_gap: int = 15
 
-    # 目标占比范围 (%)
     strand_quest_ratio_min: int = 55
     strand_quest_ratio_max: int = 65
     strand_fire_ratio_min: int = 20
@@ -164,7 +145,7 @@ class DataModulesConfig:
     strand_constellation_ratio_max: int = 20
 
     # ================= 爽点节奏 =================
-    pacing_segment_size: int = 100  # 每段分析的章节数
+    pacing_segment_size: int = 100
     pacing_words_per_point_excellent: int = 1000
     pacing_words_per_point_good: int = 1500
     pacing_words_per_point_acceptable: int = 2000
@@ -179,21 +160,17 @@ class DataModulesConfig:
         return self.webnovel_dir / "vectors.db"
 
     def ensure_dirs(self):
-        """确保必要目录存在"""
         self.webnovel_dir.mkdir(parents=True, exist_ok=True)
 
     @classmethod
     def from_project_root(cls, project_root: str | Path) -> "DataModulesConfig":
-        """从项目根目录创建配置"""
         return cls(project_root=Path(project_root))
 
 
-# 全局默认配置
 _default_config: Optional[DataModulesConfig] = None
 
 
 def get_config(project_root: Optional[Path] = None) -> DataModulesConfig:
-    """获取配置实例"""
     global _default_config
     if project_root is not None:
         return DataModulesConfig.from_project_root(project_root)
@@ -203,6 +180,5 @@ def get_config(project_root: Optional[Path] = None) -> DataModulesConfig:
 
 
 def set_project_root(project_root: str | Path):
-    """设置项目根目录"""
     global _default_config
     _default_config = DataModulesConfig.from_project_root(project_root)