API · 知识库文档上传 · 检索 · 删除知识库 API 知识库的写入与检索。 列出知识库 GET /v1/knowledge-bases?workspace_id=ws-001 上传文档 单文件 POST /v1/knowledge-bases/{id}/documents Content-Type: multipart/form-data file: <binary> group: 公开 metadata: {"author": "张三", "category": "FAQ"} 响应: { "document_id": "doc-001", "name": "FAQ.pdf", "size_bytes": 1024000, "status": "parsing", "created_at": "..." } 文本 POST /v1/knowledge-bases/{id}/documents { "type": "text", "name": "退货政策", "content": "我们的退货政策...", "metadata": { "category": "FAQ" } } 状态查询 GET /v1/documents/{document_id} { "id": "doc-001", "status": "ready", "chunks_count": 47, "indexed_at": "..." } status 取值:uploading / parsing / chunking / embedding / ready / failed。 检索 POST /v1/knowledge-bases/{id}/search { "query": "退货政策", "top_k": 5, "filters": { "category": "FAQ" }, "rerank": true } 响应: { "results": [ { "chunk_id": "ch-001", "document_id": "doc-001", "document_name": "FAQ.pdf", "content": "我们支持 7 天无理由退货...", "score": 0.87, "metadata": { "page": 3 } } ] } 删除文档 DELETE /v1/documents/{document_id} 重新解析 POST /v1/documents/{document_id}/reparse 可在更换分块策略后批量重解析。 反模式 把检索 API 当 LLM 调用 — 它只返回检索结果，不生成。要生成请用 Agent TopK 设到 100 — LLM 上下文吃不下，返回噪声反而幻觉 接下来 模型 API Webhooks 上一页API · Workflow下一页API · 模型