fix: use SiliconFlow free LLM+embedding for Graphiti

- Graphiti LLM: SiliconFlow Qwen/Qwen2.5-7B-Instruct (free, supports structured output)
- Graphiti Embedding: SiliconFlow BAAI/bge-m3 (free, OpenAI-compatible)
- MiniMax Coding Plan doesn't support structured output needed by Graphiti
- Separate GRAPHITI_LLM_* config from main LLM_* config
- Remove _wait_for_episodes call (Graphiti processes synchronously)

Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>

backend/app/api/graph.py

@@ -444,23 +444,8 @@ def build_graph():
                     progress_callback=add_progress_callback
                 )
                 
-                # 等待Zep处理完成（查询每个episode的processed状态）
-                task_manager.update_task(
-                    task_id,
-                    message=t('progress.waitingZepProcess'),
-                    progress=55
-                )
-                
-                def wait_progress_callback(msg, progress_ratio):
-                    progress = 55 + int(progress_ratio * 35)  # 55% - 90%
-                    task_manager.update_task(
-                        task_id,
-                        message=msg,
-                        progress=progress
-                    )
-                
-                builder._wait_for_episodes(episode_uuids, wait_progress_callback)
-                
+                # Graphiti 同步处理，不需要轮询 episode 状态
+
                 # 获取图谱数据
                 task_manager.update_task(
                     task_id,

backend/app/config.py

@@ -37,6 +37,16 @@ class Config:
     NEO4J_USER = os.environ.get('NEO4J_USER', 'neo4j')
     NEO4J_PASSWORD = os.environ.get('NEO4J_PASSWORD', 'foresight2026')
 
+    # Graphiti LLM 配置（SiliconFlow，用于图谱构建的 structured output）
+    GRAPHITI_LLM_API_KEY = os.environ.get('GRAPHITI_LLM_API_KEY')
+    GRAPHITI_LLM_BASE_URL = os.environ.get('GRAPHITI_LLM_BASE_URL', 'https://api.siliconflow.cn/v1')
+    GRAPHITI_LLM_MODEL = os.environ.get('GRAPHITI_LLM_MODEL', 'Qwen/Qwen2.5-7B-Instruct')
+
+    # Embedding 配置（SiliconFlow 免费 BAAI/bge-m3）
+    EMBEDDING_API_KEY = os.environ.get('EMBEDDING_API_KEY')
+    EMBEDDING_BASE_URL = os.environ.get('EMBEDDING_BASE_URL', 'https://api.siliconflow.cn/v1')
+    EMBEDDING_MODEL = os.environ.get('EMBEDDING_MODEL', 'BAAI/bge-m3')
+
     # 兼容旧配置：ZEP_API_KEY 不再需要
     ZEP_API_KEY = os.environ.get('ZEP_API_KEY', 'deprecated')
     

backend/app/services/graphiti_client.py

@@ -159,21 +159,30 @@ class GraphitiClient:
         from graphiti_core.llm_client.openai_generic_client import OpenAIGenericClient
         from graphiti_core.llm_client.config import LLMConfig
 
+        # Use SiliconFlow LLM for Graphiti (better structured output support)
+        graphiti_api_key = Config.GRAPHITI_LLM_API_KEY or Config.LLM_API_KEY
+        graphiti_base_url = Config.GRAPHITI_LLM_BASE_URL or Config.LLM_BASE_URL
+        graphiti_model = Config.GRAPHITI_LLM_MODEL or Config.LLM_MODEL_NAME
+
         llm_config = LLMConfig(
-            api_key=Config.LLM_API_KEY,
-            model=Config.LLM_MODEL_NAME,
-            small_model=Config.LLM_MODEL_NAME,
-            base_url=Config.LLM_BASE_URL,
+            api_key=graphiti_api_key,
+            model=graphiti_model,
+            small_model=graphiti_model,
+            base_url=graphiti_base_url,
         )
         llm_client = OpenAIGenericClient(config=llm_config)
 
-        # Embedder: custom MiniMax embedder (their API uses 'texts' not 'input')
-        embedder = MiniMaxEmbedder(
-            api_key=Config.LLM_API_KEY,
-            base_url=Config.LLM_BASE_URL,
+        # Embedder: SiliconFlow free BAAI/bge-m3 (OpenAI-compatible)
+        from graphiti_core.embedder.openai import OpenAIEmbedder, OpenAIEmbedderConfig
+        embedder_config = OpenAIEmbedderConfig(
+            api_key=Config.EMBEDDING_API_KEY or Config.LLM_API_KEY,
+            base_url=Config.EMBEDDING_BASE_URL or "https://api.siliconflow.cn/v1",
+            embedding_model=Config.EMBEDDING_MODEL or "BAAI/bge-m3",
+            embedding_dim=1024,
         )
+        embedder = OpenAIEmbedder(config=embedder_config)
 
-        # Reranker: use the same LLM config (MiniMax-compatible)
+        # Reranker: use the LLM config (MiniMax)
         from graphiti_core.cross_encoder.openai_reranker_client import OpenAIRerankerClient
         reranker = OpenAIRerankerClient(config=llm_config)