feat: Initialize AI generation workflow functionality

apps/models_provider/api/chat_completion.py

@@ -0,0 +1,31 @@
+# coding=utf-8
+from common.mixins.api_mixin import APIMixin
+from django.utils.translation import gettext_lazy as _
+from drf_spectacular.types import OpenApiTypes
+from drf_spectacular.utils import OpenApiParameter
+from models_provider.serializers.chat_completion_serializers import ChatCompletionSerializers
+
+
+class ChatCompletionAPI(APIMixin):
+    @staticmethod
+    def get_parameters():
+        return [
+            OpenApiParameter(
+                name="workspace_id",
+                description=_("workspace id"),
+                type=OpenApiTypes.STR,
+                location=OpenApiParameter.PATH,  # type: ignore
+                required=True,
+            ),
+            OpenApiParameter(
+                name="model_id",
+                description=_("model id"),
+                type=OpenApiTypes.STR,
+                location=OpenApiParameter.PATH,  # type: ignore
+                required=True,
+            ),
+        ]
+
+    @staticmethod
+    def get_request():
+        return ChatCompletionSerializers

apps/models_provider/serializers/chat_completion_serializers.py

@@ -0,0 +1,145 @@
+# coding=utf-8
+"""
+@project: MaxKB
+@file： chat_completion_serializers.py
+@desc: 通用流式补全（支持 function calling），供前端工作流 AI 生成的 agent 循环逐轮调用
+"""
+
+import json
+
+from common.config.embedding_config import ModelManage
+from common.exception.app_exception import AppApiException
+from common.utils.common import to_stream_response_simple
+from common.utils.logger import maxkb_logger
+from django.db.models import QuerySet
+from django.utils.translation import gettext_lazy as _
+from langchain_core.messages import convert_to_messages
+from models_provider.base_model_provider import ModelTypeConst
+from models_provider.models import Model
+from models_provider.tools import (
+    get_model,
+    get_model_default_params,
+    get_model_instance_by_model_workspace_id,
+    reset_model_params,
+)
+from rest_framework import serializers
+
+# messages 条数护栏，迭代上限由前端 agent 循环控制
+MAX_MESSAGES = 400
+
+
+class ChatCompletionSerializers(serializers.Serializer):
+    workspace_id = serializers.CharField(required=False, allow_blank=True, allow_null=True, label=_("Workspace ID"))
+    model_id = serializers.CharField(required=True, label=_("Model"))
+    model_params_setting = serializers.JSONField(required=False, default=dict, label=_("Model params setting"))
+    # OpenAI 格式消息：role 为 system/user/assistant/tool，assistant 可带 tool_calls 与 reasoning_content
+    messages = serializers.ListField(child=serializers.DictField(), required=True, label=_("Chat context"))
+    # OpenAI function 定义：[{type:'function', function:{name, description, parameters}}]
+    tools = serializers.ListField(child=serializers.DictField(), required=False, default=list, label=_("Tools"))
+
+    def is_valid(self, *, raise_exception=False):
+        super().is_valid(raise_exception=True)
+        messages = self.data.get("messages") or []
+        if not messages:
+            raise AppApiException(400, _("Messages cannot be empty"))
+        if len(messages) > MAX_MESSAGES:
+            raise AppApiException(400, _("Too many messages"))
+        for message in messages:
+            if message.get("role") not in ("system", "user", "assistant", "tool"):
+                raise AppApiException(400, _("Authentication failed. Please verify that the parameters are correct."))
+        model = QuerySet(Model).filter(id=self.data.get("model_id")).first()
+        if model is None:
+            raise AppApiException(500, _("Model does not exist"))
+        if model.model_type != ModelTypeConst.LLM.name:
+            raise AppApiException(400, _("Model does not exists or is not an LLM model"))
+        # 共享/系统级模型:workspace_id 为字符串 "None",不校验传入的工作空间;否则必须属于传入的工作空间
+        if (
+            model.workspace_id != "None"
+            and self.data.get("workspace_id")
+            and self.data.get("workspace_id") != model.workspace_id
+        ):
+            raise AppApiException(403, _("Model is not visible to the workspace"))
+        return model
+
+    def _get_model_instance(self, model):
+        model_id = self.data.get("model_id")
+        model_params_setting = self.data.get("model_params_setting") or {}
+        if model.workspace_id == "None":
+            default_model_params = get_model_default_params(model)
+            model_params = reset_model_params(default_model_params, **model_params_setting)
+            return ModelManage.get_model(model_id, lambda _id: get_model(model, **model_params))
+        return get_model_instance_by_model_workspace_id(
+            model_id=model_id, workspace_id=model.workspace_id, **model_params_setting
+        )
+
+    def chat_completion(self):
+        model = self.is_valid(raise_exception=True)
+        messages = convert_to_messages([dict(m) for m in self.data.get("messages")])
+        tools = [dict(t) for t in self.data.get("tools") or []]
+
+        def event(payload: dict) -> str:
+            return "data: " + json.dumps(payload, ensure_ascii=False) + "\n\n"
+
+        def process():
+            try:
+                model_instance = self._get_model_instance(model)
+                llm = model_instance.bind_tools(tools) if tools else model_instance
+                content = ""
+                reasoning = ""
+                finish_reason = None
+                usage = None
+                # 按流式 index 聚合工具调用：id/name 只在首片出现，arguments 逐片拼接
+                tool_calls_by_index: dict = {}
+                for chunk in llm.stream(messages):
+                    text = chunk.content if isinstance(chunk.content, str) else ""
+                    if text:
+                        content += text
+                        yield event({"type": "delta", "content": text})
+                    reasoning_delta = (chunk.additional_kwargs or {}).get("reasoning_content")
+                    if reasoning_delta:
+                        reasoning += reasoning_delta
+                        yield event({"type": "reasoning", "content": reasoning_delta})
+                    for tool_chunk in getattr(chunk, "tool_call_chunks", None) or []:
+                        index = tool_chunk.get("index") or 0
+                        call = tool_calls_by_index.setdefault(index, {"id": "", "name": "", "arguments": ""})
+                        if tool_chunk.get("id"):
+                            call["id"] = tool_chunk["id"]
+                        if tool_chunk.get("name"):
+                            call["name"] = call["name"] or tool_chunk["name"]
+                        args_delta = tool_chunk.get("args") or ""
+                        call["arguments"] += args_delta
+                        if call["id"]:
+                            yield event(
+                                {"type": "tool_call", "id": call["id"], "name": call["name"], "arguments": args_delta}
+                            )
+                    metadata = chunk.response_metadata or {}
+                    if metadata.get("finish_reason"):
+                        finish_reason = metadata["finish_reason"]
+                    if getattr(chunk, "usage_metadata", None):
+                        usage = dict(chunk.usage_metadata)
+                tool_calls = [
+                    {
+                        "id": call["id"],
+                        "type": "function",
+                        "function": {"name": call["name"], "arguments": call["arguments"] or "{}"},
+                    }
+                    for _index, call in sorted(tool_calls_by_index.items())
+                    if call["id"] and call["name"]
+                ]
+                yield event(
+                    {
+                        "type": "completion",
+                        "content": content,
+                        "reasoningKey": "reasoning_content" if reasoning else None,
+                        "reasoningContent": reasoning or None,
+                        "toolCalls": tool_calls,
+                        "finishReason": finish_reason or ("tool_calls" if tool_calls else "stop"),
+                        "usage": usage,
+                    }
+                )
+            except Exception as e:
+                maxkb_logger.error(f"chat completion error: {e}", exc_info=True)
+                yield event({"type": "error", "message": str(e)})
+            yield "data: [DONE]\n\n"
+
+        return to_stream_response_simple(process())

apps/models_provider/urls.py

@@ -17,6 +17,7 @@
     path('workspace/<str:workspace_id>/model_list', views.ModelList.as_view()),
     path('workspace/<str:workspace_id>/model/<str:model_id>/model_params_form', views.ModelSetting.ModelParamsForm.as_view()),
     path('workspace/<str:workspace_id>/model/<str:model_id>/prompt_generate', views.ModelPromptGenerateView.as_view()),
+    path('workspace/<str:workspace_id>/model/<str:model_id>/chat_completion', views.ModelChatCompletionView.as_view()),
     path('workspace/<str:workspace_id>/model/<str:model_id>', views.ModelSetting.Operate.as_view()),
     path('workspace/<str:workspace_id>/model/<str:model_id>/pause_download', views.ModelSetting.PauseDownload.as_view()),
     path('workspace/<str:workspace_id>/model/<str:model_id>/meta', views.ModelSetting.ModelMeta.as_view()),

apps/models_provider/views/model.py

@@ -20,9 +20,11 @@
 from django.utils.translation import gettext_lazy as _
 from drf_spectacular.utils import extend_schema
 from models_provider.api.model import DefaultModelResponse, GetModelApi, ModelCreateAPI, ModelEditApi, ModelListResponse
+from models_provider.api.chat_completion import ChatCompletionAPI
 from models_provider.api.prompt_generate import PromptGenerateAPI
 from models_provider.api.provide import ProvideApi
 from models_provider.models import Model
+from models_provider.serializers.chat_completion_serializers import ChatCompletionSerializers
 from models_provider.serializers.model_serializer import ModelSerializer, WorkspaceSharedModelSerializer
 from models_provider.serializers.prompt_generate_serializers import PromptGenerateSerializers
 from rest_framework.request import Request
@@ -413,3 +415,29 @@ def post(self, request: Request, workspace_id: str, model_id: str):
         return PromptGenerateSerializers(
             data={**request.data, "workspace_id": workspace_id, "model_id": model_id}
         ).generate_prompt()
+
+
+class ModelChatCompletionView(APIView):
+    authentication_classes = [TokenAuth]
+
+    @extend_schema(
+        methods=["POST"],
+        summary=_("Chat completion"),
+        description=_("Stream chat completion with function calling"),
+        operation_id=_("Chat completion"),  # type: ignore
+        parameters=ChatCompletionAPI.get_parameters(),
+        request=ChatCompletionAPI.get_request(),
+        responses=None,
+        tags=[_("Model")],  # type: ignore
+    )
+    @has_permissions(
+        PermissionConstants.MODEL_READ.get_workspace_model_permission(),
+        PermissionConstants.MODEL_READ.get_workspace_permission_workspace_manage_role(),
+        RoleConstants.WORKSPACE_MANAGE.get_workspace_role(),
+        RoleConstants.USER.get_workspace_role(),
+    )
+    def post(self, request: Request, workspace_id: str, model_id: str):
+        # agent 循环每轮都会调用，不记操作日志
+        return ChatCompletionSerializers(
+            data={**request.data, "workspace_id": workspace_id, "model_id": model_id}
+        ).chat_completion()

ui/src/api/API_README.md

@@ -193,7 +193,8 @@ Action、Drawer 或 Dialog。复用方直接使用 `typeof XxxApi` 约束完整
 智能体、工具、知识库 Workflow API 分别维护 application_version、tool_version、knowledge_version，
 使用同名 getWorkflowVersions／putWorkflowVersion，类型为 WorkflowVersion／WorkflowVersionPayload。
 前端版本标题上限 64、更新说明上限 1000，提交字段为 name、publish_desc，
-不将资源 desc 当作版本说明。智能体后端已支持 publish_desc；工具和知识库版本当前只处理 name。
+不将资源 desc 当作版本说明。智能体和工具后端已支持 publish_desc；知识库版本当前只处理 name。
+智能体与工具的 putXxxPublish 提交 publish_name、publish_desc，由发布弹框填写。
 三个版本列表均未返回历史默认模型配置，预览与恢复保留当前配置。
 
 `admin/store.ts` 的 getStoreApplicationList、getStoreToolWorkflowList、getStoreKnowledgeList 查询三类模板，
@@ -269,6 +270,16 @@ System 接口不依赖工作空间 ID。
 `useWorkflowStore(resourceScope).postPromptGenerate` 调用；三个范围的 Store 适配器分别引用对应的
 Model API，直接转发生成请求，不使用查询缓存或请求去重。
 
+### 流式补全
+
+Workspace、System 资源管理和 System 共享资源均在对应范围的 `model.ts` 提供
+`postChatCompletion(modelId, payload, signal)`，请求结构为 `ChatCompletionPayload`（OpenAI 格式的 messages 与 tools）。
+路径为 `/workspace/<workspaceId>/model/<modelId>/chat_completion`、`/system/resource/model/<modelId>/chat_completion` 与
+`/system/shared/model/<modelId>/chat_completion`；当前后端仅实现 Workspace 接口，System 接口需部署提供对应扩展。
+均返回原始 SSE Response，事件为 `delta`、`reasoning`、`tool_call`、`completion` 与 `error`；`signal` 用于中止在途请求，
+`postStream` 的第四个可选参数透传给 `fetch`。工作流 AI 生成通过 `useWorkflowStore(resourceScope).postChatCompletion` 调用，
+Store 直接转发，不缓存或去重。
+
 ### 知识库维护
 
 Workspace、System 资源管理、System 共享各自的 knowledge.ts 保留真实能力，Workspace 共享只查询。

ui/src/api/admin/core/request.ts

@@ -192,7 +192,7 @@ export async function downloadRequest(url: string, method: string, data?: unknow
 }
 
 /** 发送 POST 请求并返回可逐块读取的原始响应。 */
-export function postStream(base: string, path: string, data?: unknown): Promise<Response> {
+export function postStream(base: string, path: string, data?: unknown, signal?: AbortSignal): Promise<Response> {
   const { auth, user } = useStore()
   const headers: Record<string, string> = { 'Content-Type': 'application/json' }
   if (auth.token) {
@@ -205,6 +205,7 @@ export function postStream(base: string, path: string, data?: unknown): Promise<
     method: 'POST',
     headers,
     body: data === undefined ? undefined : JSON.stringify(data),
+    signal,
   })
 }
 

ui/src/api/admin/system/resource-management/model.ts

@@ -1,6 +1,6 @@
 import { del, get, post, postStream, put } from '../../core/request'
 import type { ParamsPage, ResponsePage } from '../../core/types'
-import type { Dict, DynamicFormField, ModelItem, ModelPayload, PromptGeneratePayload } from '@/api/types'
+import type { Dict, DynamicFormField, ModelItem, ModelPayload, ChatCompletionPayload, PromptGeneratePayload } from '@/api/types'
 import { ADMIN_API_BASE_PATH } from '@/api/constants'
 
 const prefix = '/system/resource/model'
@@ -35,10 +35,25 @@ const getModelParamsForm = (modelId: string) => get<DynamicFormField[]>(`${prefi
 const putModelParamsForm = (modelId: string, payload: DynamicFormField[]) =>
   put<DynamicFormField[], boolean>(`${prefix}/${modelId}/model_params_form`, payload)
 
-
 /** 使用指定模型流式生成或优化系统提示词。 */
 const postPromptGenerate = (modelId: string, payload: PromptGeneratePayload) => {
   return postStream(ADMIN_API_BASE_PATH, `/system/resource/model/${modelId}/prompt_generate`, { ...payload })
 }
 
-export default { getModelPage, getModelListWithShared, getModelDetail, postModel, putModel, deleteModel, getModelParamsForm, putModelParamsForm, postPromptGenerate }
+/** 流式补全（支持 function calling），返回原始 SSE Response；signal 用于中止在途请求。 */
+const postChatCompletion = (modelId: string, payload: ChatCompletionPayload, signal?: AbortSignal) => {
+  return postStream(ADMIN_API_BASE_PATH, `/system/resource/model/${modelId}/chat_completion`, { ...payload }, signal)
+}
+
+export default {
+  getModelPage,
+  getModelListWithShared,
+  getModelDetail,
+  postModel,
+  putModel,
+  deleteModel,
+  getModelParamsForm,
+  putModelParamsForm,
+  postPromptGenerate,
+  postChatCompletion,
+}

ui/src/api/admin/system/resource-management/tool/tool-workflow.ts

@@ -29,9 +29,12 @@ const putToolWorkflow = (toolId: string, payload: ToolWorkflowPayload) => {
   return put<ToolWorkflowPayload, ToolWorkflowDetail>(`${prefix}/${toolId}/workflow`, payload)
 }
 
-/** 发布工具工作流。 */
-const putToolWorkflowPublish = (toolId: string) => {
-  return put<undefined, boolean>(`${prefix}/${toolId}/publish`)
+/** 发布工具工作流，并记录版本标题和更新说明。 */
+const putToolWorkflowPublish = (toolId: string, publishName: string, publishDesc?: string) => {
+  return put<Record<string, unknown>, boolean>(`${prefix}/${toolId}/publish`, {
+    publish_name: publishName,
+    publish_desc: publishDesc,
+  })
 }
 
 /** 调试已保存的工具工作流，返回 SSE 响应。 */

ui/src/api/admin/system/shared-resources/model.ts

@@ -1,5 +1,5 @@
 import { del, get, post, postStream, put } from '../../core/request'
-import type { Dict, DynamicFormField, ModelItem, ModelPayload, PromptGeneratePayload } from '@/api/types'
+import type { Dict, DynamicFormField, ModelItem, ModelPayload, ChatCompletionPayload, PromptGeneratePayload } from '@/api/types'
 import { ADMIN_API_BASE_PATH } from '@/api/constants'
 
 const prefix = '/system/shared/model'
@@ -61,6 +61,11 @@ const postPromptGenerate = (modelId: string, payload: PromptGeneratePayload) =>
   return postStream(ADMIN_API_BASE_PATH, `/system/shared/model/${modelId}/prompt_generate`, { ...payload })
 }
 
+/** 流式补全（支持 function calling），返回原始 SSE Response；signal 用于中止在途请求。 */
+const postChatCompletion = (modelId: string, payload: ChatCompletionPayload, signal?: AbortSignal) => {
+  return postStream(ADMIN_API_BASE_PATH, `/system/shared/model/${modelId}/chat_completion`, { ...payload }, signal)
+}
+
 export default {
   deleteModel,
   getModelDetail,
@@ -70,6 +75,7 @@ export default {
   getModelParamsForm,
   postModel,
   postPromptGenerate,
+  postChatCompletion,
   putModel,
   putModelParamsForm,
   putPauseModelDownload,

ui/src/api/admin/system/shared-resources/tool/tool-workflow.ts

@@ -29,9 +29,12 @@ const putToolWorkflow = (toolId: string, payload: ToolWorkflowPayload) => {
   return put<ToolWorkflowPayload, ToolWorkflowDetail>(`${prefix}/${toolId}/workflow`, payload)
 }
 
-/** 发布工具工作流。 */
-const putToolWorkflowPublish = (toolId: string) => {
-  return put<undefined, boolean>(`${prefix}/${toolId}/publish`)
+/** 发布工具工作流，并记录版本标题和更新说明。 */
+const putToolWorkflowPublish = (toolId: string, publishName: string, publishDesc?: string) => {
+  return put<Record<string, unknown>, boolean>(`${prefix}/${toolId}/publish`, {
+    publish_name: publishName,
+    publish_desc: publishDesc,
+  })
 }
 
 /** 调试已保存的工具工作流，返回 SSE 响应。 */

ui/src/api/admin/workspace/model.ts

@@ -1,7 +1,7 @@
 import { del, get, post, postStream, put } from '../core/request'
 import type { Dict, DynamicFormField, ModelPayload, ModelItem } from '@/api/types'
 import { getWorkspaceId } from '@/utils/resource-context'
-import type { PromptGeneratePayload } from '@/api/types'
+import type { ChatCompletionPayload, PromptGeneratePayload } from '@/api/types'
 import { ADMIN_API_BASE_PATH } from '@/api/constants'
 
 const getPrefix = () => {
@@ -72,13 +72,19 @@ const postPromptGenerate = (modelId: string, payload: PromptGeneratePayload) =>
   })
 }
 
+/** 流式补全（支持 function calling），返回原始 SSE Response；signal 用于中止在途请求。 */
+const postChatCompletion = (modelId: string, payload: ChatCompletionPayload, signal?: AbortSignal) => {
+  return postStream(ADMIN_API_BASE_PATH, `/workspace/${getWorkspaceId()}/model/${modelId}/chat_completion`, { ...payload }, signal)
+}
+
 export default {
   deleteModel,
   getModelDetail,
   getModelList,
   getModelListWithShared,
   getModelMeta,
   getModelParamsForm,
+  postChatCompletion,
   postModel,
   postPromptGenerate,
   putModel,