feat(api): 接入智谱 embedding-3（1024 维）并迁移 memory_chunks 向量列

api/app/adapters/embedding/openai.py

@@ -1,21 +0,0 @@
-"""OpenAI embedding adapter — implements EmbeddingProvider port."""
-
-from openai import AsyncOpenAI
-
-
-class OpenAIEmbeddingProvider:
-    def __init__(self, api_key: str, model: str = "text-embedding-3-small"):
-        self._client = AsyncOpenAI(api_key=api_key) if api_key else None
-        self._model = model
-
-    async def embed_text(self, text: str) -> list[float]:
-        if not self._client:
-            return []
-        resp = await self._client.embeddings.create(input=[text], model=self._model)
-        return resp.data[0].embedding
-
-    async def embed_texts(self, texts: list[str]) -> list[list[float]]:
-        if not self._client or not texts:
-            return []
-        resp = await self._client.embeddings.create(input=texts, model=self._model)
-        return [item.embedding for item in sorted(resp.data, key=lambda d: d.index)]

api/app/adapters/embedding/zhipu.py

@@ -0,0 +1,56 @@
+"""智谱 BigModel 国内 embedding API — 实现 EmbeddingProvider（zai-sdk / ZhipuAiClient）。"""
+
+from __future__ import annotations
+
+import asyncio
+
+from zai import ZhipuAiClient
+
+from app.core.embedding import MEMORY_EMBEDDING_DIMENSION
+
+# 单次请求最多 64 条文本（智谱 Embedding-3 文档）
+_EMBED_BATCH_SIZE = 64
+
+
+class ZhipuEmbeddingProvider:
+    def __init__(
+        self,
+        *,
+        api_key: str,
+        base_url: str | None = None,
+        model: str = "embedding-3",
+    ) -> None:
+        self._model = model
+        if not api_key:
+            self._client = None
+        elif base_url:
+            self._client = ZhipuAiClient(
+                api_key=api_key,
+                base_url=base_url.rstrip("/"),
+            )
+        else:
+            self._client = ZhipuAiClient(api_key=api_key)
+
+    def _create_vectors_sync(self, texts: list[str]) -> list[list[float]]:
+        assert self._client is not None
+        resp = self._client.embeddings.create(
+            input=texts,
+            model=self._model,
+            dimensions=MEMORY_EMBEDDING_DIMENSION,
+        )
+        ordered = sorted(resp.data, key=lambda d: d.index or 0)
+        return [list(item.embedding) for item in ordered]
+
+    async def embed_text(self, text: str) -> list[float]:
+        vectors = await self.embed_texts([text])
+        return vectors[0] if vectors else []
+
+    async def embed_texts(self, texts: list[str]) -> list[list[float]]:
+        if not self._client or not texts:
+            return []
+        out: list[list[float]] = []
+        for i in range(0, len(texts), _EMBED_BATCH_SIZE):
+            batch = texts[i : i + _EMBED_BATCH_SIZE]
+            part = await asyncio.to_thread(self._create_vectors_sync, batch)
+            out.extend(part)
+        return out

api/app/core/config.py

@@ -49,6 +49,11 @@ class Settings(BaseSettings):
     llm_model: str = ""
     llm_temperature: float = 0.7
 
+    # ── 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（短回复：token 上限 + 代码截断，见 reply_limits）──
     chat_interview_max_tokens: int = 320
     chat_interview_max_segments: int = 2

api/app/core/dependencies.py

@@ -124,10 +124,13 @@ def get_object_storage() -> ObjectStorage:
 
 @lru_cache
 def get_embedding_provider() -> EmbeddingProvider:
-    from app.adapters.embedding.openai import OpenAIEmbeddingProvider
+    from app.adapters.embedding.zhipu import ZhipuEmbeddingProvider
 
-    api_key = settings.openai_api_key or settings.deepseek_api_key
-    return OpenAIEmbeddingProvider(api_key=api_key)
+    return ZhipuEmbeddingProvider(
+        api_key=settings.zhipu_api_key,
+        base_url=settings.embedding_base_url or None,
+        model=settings.embedding_model,
+    )
 
 
 # ── Auth dependencies ────────────────────────────────────────

api/app/core/embedding.py

@@ -0,0 +1,6 @@
+"""Memory chunk 向量维度（与智谱 embedding-3、pgvector 列一致）。
+
+本期固定 1024；若调整维度需独立迁移与排期，勿仅改此处常量。
+"""
+
+MEMORY_EMBEDDING_DIMENSION = 1024

api/app/features/memory/models.py

@@ -10,12 +10,13 @@ from sqlalchemy import (
     String,
     Text,
 )
-from sqlalchemy.orm import relationship
 from sqlalchemy.dialects.postgresql import TSVECTOR as TSVector
+from sqlalchemy.orm import relationship
 
 from app.core.db import Base, utc_now
+from app.core.embedding import MEMORY_EMBEDDING_DIMENSION
 
-pgvector_type = Vector(1536)
+pgvector_type = Vector(MEMORY_EMBEDDING_DIMENSION)
 
 
 class MemorySource(Base):