EC-RAG 26.01 release with code enhancement and bug fix

.github/workflows/scripts/get_test_matrix.sh

@@ -26,7 +26,7 @@ for example in ${examples}; do
     echo -e "Test supported hardware list: \n${hardware_list}"
 
     run_hardware=""
-    if [[ $(printf '%s\n' "${changed_files[@]}" | grep ${example} | cut -d'/' -f2 | grep -E '\.py|Dockerfile*|ui|docker_image_build' ) ]]; then
+    if [[ $(printf '%s\n' "${changed_files[@]}" | grep ${example} | cut -d'/' -f2 | grep -E '\.py|Dockerfile*|ui|edgecraftrag|docker_image_build' ) ]]; then
         echo "run test on all hardware if megaservice or ui code change..."
         run_hardware=$hardware_list
     elif [[ $(printf '%s\n' "${changed_files[@]}" | grep ${example} | grep 'tests'| cut -d'/' -f3 | grep -vE '^test_|^_test' ) ]]; then

EdgeCraftRAG/docker_compose/intel/gpu/arc/compose.yaml

@@ -139,7 +139,7 @@ services:
     ports:
       - ${VLLM_SERVICE_PORT_B60:-8086}:${VLLM_SERVICE_PORT_B60:-8086}
     volumes:
-      - ${MODEL_PATH}:/workspace/vllm/models
+      - ${MODEL_PATH-${PWD}}:/workspace/vllm/models
     devices:
       - /dev/dri:/dev/dri
     environment:

EdgeCraftRAG/docs/API_Guide.md

@@ -60,7 +60,7 @@ Benchmarking activities may significantly reduce system performance.
 export ENABLE_BENCHMARK="true"
 
 # check the benchmark data for pipeline {pipeline_name}
-curl -X GET http://${HOST_IP}:16010/v1/settings/pipelines/{pipeline_name}/benchmark -H "Content-Type: application/json" | jq '.'
+curl -X GET http://${HOST_IP}:16010/v1/settings/pipelines/{pipeline_name}/benchmarks -H "Content-Type: application/json" | jq '.'
 ```
 
 ## Model Management

EdgeCraftRAG/docs/Agentic_RAG_Guide_Zh.md

@@ -11,7 +11,7 @@ EC-RAG目前支持两种类型的智能体：
 - **Simple Agent**: 不包含规划、research的简单轻量级Agent；
 - **Deep Search Agent**: 能够进行多步规划、deep research和深度推理的更高级版本Agent。
 
-## Retrieval
+## 创建智能体
 
 ### 前提条件
 

EdgeCraftRAG/docs/Experience_Guide.md

@@ -0,0 +1,55 @@
+# EC-RAG Experience Guide
+
+This guide explains how to enable, configure, and use Experience capabilities within EdgeCraftRAG (EC-RAG). It covers the creation and management of experience to enhance the answer quality
+
+## Overview
+
+Experience refers to the knowledge and skills acquired through practical involvement, trial, and reflection, serving as a key foundation for solving real-world problems. In EC-RAG we now provide abilities to add such experience for the reference of your Chat.
+
+Before use Experience, please make sure Milvus VectorDB is enabled and below env is set:
+
+```bash
+export METADATA_DATABASE_URL="http://${HOST_IP}:19530"
+```
+
+## Creating An Experience
+
+Upon successful deployment of EC-RAG, we may navigate to `Knowledge Base` tab from the left panel:
+![alt text](../assets/img/experience_access.png)
+
+Then in the `Create Type Select` windows, choose `Go Create Experience`:
+![alt text](../assets/img/experience_create_1.png)
+
+In the next windws, we can set Experience to `Activated` or `Inactive`, this could also be changed after creation of Experience:
+![alt text](../assets/img/experience_create_2.png)
+
+## Managing Experiences
+
+### Add Experiences
+
+After successfully created `Experience`, Experience tab would be available with `Unique` tag:
+![alt text](../assets/img/experience_management_1.png)
+
+To add new experience, select `Create Experience` from the right panel:
+![alt text](../assets/img/experience_management_2.png)
+
+In the `Create Experience` window, we could add experience and its corresponding content as needed, multi experience creation is supported:
+![alt text](../assets/img/experience_management_3.png)
+
+### View and Edit Experiences
+
+After experience added, they will be shown in list under `Experience` tab. We could toggle the `+` sign to show experience content details. Use the three dots next to `Experience` to activate or deactivate a Experience Base:
+![alt text](../assets/img/experience_management_4.png)
+
+## Examples
+
+Here's a comparison with experience activated and not activated:
+
+Without experience:
+![alt text](../assets/img/experience_example_3.png)
+
+Activated experience:
+![alt text](../assets/img/experience_example_1.png)
+
+RAG Answer:
+![alt text](../assets/img/experience_example_2.png)

EdgeCraftRAG/docs/Experience_Guide_Zh.md

@@ -0,0 +1,54 @@
+# EC-RAG 经验注入指南
+
+本指南将为您介绍如何在 EdgeCraftRAG (EC-RAG) 中启用、配置和使用经验注入（Experience）功能，涵盖 Experience 的创建和管理，旨在提升您使用EC-RAG时的回答质量。
+
+## 概述
+
+Experience（经验）是指个人或团队在实践过程中积累的知识和技能，通常通过实际操作、试错和反思获得，是解决实际问题的重要依据。EC-RAG 现已支持添加此类Experience，为您的对话提供更多参考。
+使用Experience之前，请确保Mivlus服务已经启动，并且请指定如下的环境变量:
+
+```bash
+export METADATA_DATABASE_URL="http://${HOST_IP}:19530"
+```
+
+## 创建 Experience
+
+当 EC-RAG 被正确部署后，您可以通过左侧面板的`知识库`（Knowledge Base）选项卡切换到Experience创建页面：
+![alt text](../assets/img/experience_access_zh.png)
+
+然后在`创建类型选择`窗口中，选择`去创建经验注入`：
+![alt text](../assets/img/experience_create_1_zh.png)
+
+在接下来的窗口中，您可以将 Experience 设置为`已启用`或`已停用`状态。请注意，这也可以在创建 Experience 后进行更改：
+![alt text](../assets/img/experience_create_2_zh.png)
+
+## 管理 Experience
+
+成功创建`Experience`后，左侧面板会出现`Experience`选项卡，同时带有`唯一`标签：
+![alt text](../assets/img/experience_management_1_zh.png)
+
+### 添加 Experience
+
+当您需要添加新 Experience 时，请从右侧面板选择`新建经验`：
+![alt text](../assets/img/experience_management_2_zh.png)
+
+在`新建经验`窗口中，您可以根据需要添加经验及其对应的内容，EC-RAG 支持创建多个经验条目：
+![alt text](../assets/img/experience_management_3_zh.png)
+
+### 查看与修改
+
+添加经验后，它们将以列表形式显示在`经验`选项卡下。您可以点击`+`号展开查看 Experience 内容的详细信息。使用左侧面板里`Experience`旁边的三个点可以激活或停用某个 Experience 库：
+![alt text](../assets/img/experience_management_4_zh.png)
+
+## 使用示例
+
+以下是激活和未激活Experience库的对比：
+
+未激活Experience：
+![alt text](../assets/img/experience_example_3.png)
+
+激活Experience：
+![alt text](../assets/img/experience_example_1.png)
+
+RAG回答：
+![alt text](../assets/img/experience_example_2.png)

EdgeCraftRAG/edgecraftrag/VERSION

@@ -1 +1 @@
-26.01-Dev
+26.01-Release

EdgeCraftRAG/edgecraftrag/api/v1/agent.py

@@ -5,6 +5,7 @@
 import os
 import time
 
+from edgecraftrag.api.v1.knowledge_base import Synchronizing_vector_data
 from edgecraftrag.api_schema import AgentCreateIn
 from edgecraftrag.base import AgentType
 from edgecraftrag.config_repository import MilvusConfigRepository, save_agent_configurations
@@ -22,12 +23,14 @@ async def get_all_agents():
     agents = ctx.get_agent_mgr().get_agents()
     active_id = ctx.get_agent_mgr().get_active_agent_id()
     for k, agent in agents.items():
+        pipeline_name = ctx.get_pipeline_mgr().get_pipeline_by_name_or_id(agent.pipeline_idx).name
         out.append(
             AgentCreateIn(
                 idx=agent.idx,
                 name=agent.name,
                 type=agent.comp_subtype,
                 pipeline_idx=agent.pipeline_idx,
+                pipeline_name=pipeline_name,
                 configs=agent.configs,
                 active=True if agent.idx == active_id else False,
             )
@@ -40,12 +43,14 @@ async def get_all_agents():
 async def get_agent(name):
     agent = ctx.get_agent_mgr().get_agent_by_name(name)
     if agent:
+        pipeline_name = ctx.get_pipeline_mgr().get_pipeline_by_name_or_id(agent.pipeline_idx).name
         isactive = True if agent.idx == ctx.get_agent_mgr().get_active_agent_id() else False
         return AgentCreateIn(
             idx=agent.idx,
             name=agent.name,
             type=agent.comp_subtype,
             pipeline_idx=agent.pipeline_idx,
+            pipeline_name=pipeline_name,
             configs=agent.configs,
             active=isactive,
         )
@@ -70,10 +75,15 @@ async def create_agent(request: AgentCreateIn, status_code=status.HTTP_201_CREAT
 async def update_agent(name, request: AgentCreateIn):
     try:
         agentmgr = ctx.get_agent_mgr()
-        if agentmgr.get_agent_by_name(name):
+        agent = agentmgr.get_agent_by_name(name)
+        if agent:
             ret = agentmgr.update_agent(name, request)
             if ret:
                 await save_agent_configurations("update", ctx.get_agent_mgr().get_agents())
+                # manage agent bound pipeline status, trigger kb indexing if needed
+                # can be removed once kb indexing is decoupled from pipeline
+                pl_idx = agent.pipeline_idx
+                await manage_agent_bound_pipeline(pl_idx, request)
             return ret
         else:
             raise HTTPException(status_code=status.HTTP_404_NOT_FOUND)
@@ -146,3 +156,36 @@ async def load_agent(request: AgentCreateIn):
         agentmgr.remove_agent_by_name(request.name)
         raise HTTPException(status_code=status.HTTP_500_INTERNAL_SERVER_ERROR, detail=str(e))
     return agent
+
+
+async def manage_agent_bound_pipeline(bound_pl_idx, request):
+    # case1: activate agent, while bound pipeline is not active -> activate it, cache previous active pipeline if exists
+    # case2: activate agent, while bound pipeline is already active -> still call activate, for caching current pipeline
+    # case3: deactivate agent, while bound pipeline **was** active -> do NOT deactivate bound pipeline, do nothing
+    # case4: deactivate agent, while bound pipeline **was NOT** active -> deactivate bound pipeline, activate previous active pipeline if exists
+    pl_manager = ctx.get_pipeline_mgr()
+    prev_active_pipeline = pl_manager.get_active_pipeline()
+
+    active_kb = ctx.knowledgemgr.get_active_knowledge_base()
+    kb_name = active_kb.name if active_kb else "default"
+
+    if request.active:
+        pl_manager.activate_pipeline(bound_pl_idx, request.active, ctx.get_node_mgr(), kb_name, cache_prev=True)
+    else:
+        # at deactivate, prev_active_pl can be 1.other pl/2.None/3.current bound_pl
+        prev_active_pl = pl_manager.get_prev_active_pipeline_name()
+        if prev_active_pl and prev_active_pl != bound_pl_idx:
+            # 1, restore to the other pipeline activated
+            pl_manager.activate_pipeline(prev_active_pl, True, ctx.get_node_mgr(), kb_name)
+        elif not prev_active_pl:
+            # 2, deactivate current bound pipeline, leave no active pipeline as before
+            pl_manager.activate_pipeline(bound_pl_idx, False, ctx.get_node_mgr(), kb_name)
+        else:
+            # 3, do nothing
+            pass
+        # when agent is deactivated, clear cached previous active pipeline
+        pl_manager.clear_prev_active_pipeline_name()
+
+    cur_active_pipeline = pl_manager.get_active_pipeline()
+    if prev_active_pipeline and cur_active_pipeline and prev_active_pipeline.idx != cur_active_pipeline.idx:
+        await Synchronizing_vector_data(prev_active_pipeline, cur_active_pipeline)

EdgeCraftRAG/edgecraftrag/api/v1/chatqna.py

@@ -9,6 +9,7 @@
 import requests
 from comps.cores.proto.api_protocol import ChatCompletionRequest
 from edgecraftrag.api_schema import RagOut
+from edgecraftrag.base import GeneratorType
 from edgecraftrag.context import ctx
 from edgecraftrag.utils import chain_async_generators, serialize_contexts, stream_generator
 from fastapi import Body, FastAPI, HTTPException, status
@@ -52,19 +53,21 @@ async def chatqna(request: ChatCompletionRequest):
         if experience_kb:
             request.tool_choice = "auto" if experience_kb.experience_active else "none"
 
-        request.input = ctx.get_session_mgr().concat_history(
-            sessionid, active_pl.generator.inference_type, request.messages
-        )
+        generator = active_pl.get_generator(GeneratorType.CHATQNA)
+        inference_type = generator.inference_type if generator else "local"
+
+        request.input = ctx.get_session_mgr().concat_history(sessionid, inference_type, request.messages)
 
         # Run agent if activated, otherwise, run pipeline
         if ctx.get_agent_mgr().get_active_agent():
             run_agent_gen = await ctx.get_agent_mgr().run_agent(chat_request=request)
             return StreamingResponse(save_session(sessionid, run_agent_gen), media_type="text/plain")
 
         else:
-            generator = active_pl.generator
-            if generator:
-                request.model = generator.model_id
+            generator = active_pl.get_generator(GeneratorType.CHATQNA)
+            if not generator:
+                raise Exception("code:0000Please make sure chatqna generator is available in pipeline.")
+            request.model = generator.model_id
 
         if request.stream:
             run_pipeline_gen, contexts = await ctx.get_pipeline_mgr().run_pipeline(chat_request=request)
@@ -75,6 +78,8 @@ async def chatqna(request: ChatCompletionRequest):
             return str(ret)
 
     except Exception as e:
+        if "code:0000" in str(e):
+            return str(e)
         raise HTTPException(
             status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
             detail=f"ChatQnA Error: {e}",
@@ -91,7 +96,7 @@ async def ragqna(request: ChatCompletionRequest):
         request.user = active_kb if active_kb else None
         if experience_kb:
             request.tool_choice = "auto" if experience_kb.experience_active else "none"
-        generator = ctx.get_pipeline_mgr().get_active_pipeline().generator
+        generator = ctx.get_pipeline_mgr().get_active_pipeline().get_generator(GeneratorType.CHATQNA)
         if generator:
             request.model = generator.model_id
         if request.stream: