Elasticsearch:如何使用 Connectors 接入数据到 Elastic Agent builder 里进行企业搜索

0 阅读5分钟

在之前的文章 “Kibana 中的人工智能连接器:你的 agent 可调用的 90 多个数据源”,Kibana 提供了超过 90 多个连接器供我们的 Agents 来使用。我们可以很方便地把自己的企业数据接入到 agents 里,并对它们查询使用。Kibana 内置了超过 90 个连接器规范,涵盖 Gmail、谷歌云端硬盘、谷歌日历、Slack、 GitHub 、Confluence、PagerDuty、Jira、Salesforce、Datadog 和 Snowflake:

在今天的文章里,我们来详细介绍如何使用 github 连接器接入 github issue 的数据,并供我们查询。

步骤一:申请 github API key

我们可以 github 的图像处找到  Settings,并获取 API key:

最终,我们得到一个属于我们自己的 API token 。

步骤二:创建一个 github 连接器

我们在 Kibana 中创建一个 github 连接器:

这样,我们就创建了一个叫做 my-github1 ID 的 github 连接器。

步骤三:使用上面的 github 连接器创建一个 workflow

我们使用上面创建的 github ID 创建一个如下的 workflow。它用来读取所有的 issues:

`

1.  version: "1"
2.  name: GitHub Issue Analyzer1
3.  enabled: true
4.  consts:
5.    repo_owner: liu-xiao-guo
6.    repo_name: elastic_multimodal_search
7.  triggers:
8.    - type: manual
9.  steps:
10.    - name: search_issue
11.      type: github.searchIssues
12.      with:
13.        query: repo:liu-xiao-guo/elastic_multimodal_search is:issue
14.        order: asc
15.        sort: created
16.        page: 1
17.        perPage: 100
18.      connector-id: my-github1
19.    - name: analyze_issues
20.      type: foreach
21.      foreach: "{{ steps.search_issue.output.items | json }}"
22.      steps:
23.        - name: extract_issue
24.          type: data.set
25.          with:
26.            title: "{{ foreach.item.title }}"
27.            body: "{{ foreach.item.body }}"
28.            html_url: "{{ foreach.item.html_url }}"
29.            state: "{{ foreach.item.state }}"
30.            number: "{{ foreach.item.number }}"
31.        - name: get_issue_comments
32.          type: github.callTool
33.          with:
34.            name: issue_read
35.            arguments:
36.              method: get_comments
37.              owner: "{{ consts.repo_owner }}"
38.              repo: "{{ consts.repo_name }}"
39.              issue_number: "{{ foreach.item.number | plus: 0 }}"
40.          connector-id: my-github1
41.        - name: analyze_issue
42.          type: ai.prompt
43.          with:
44.            prompt: |
45.              Please analyze the following GitHub issue and suggest a concrete solution.

47.              **Issue #{{ steps.extract_issue.output.number }}**: {{ steps.extract_issue.output.title }}

49.              **State**: {{ steps.extract_issue.output.state }}

51.              **Description**:
52.              {{ steps.extract_issue.output.body }}

54.              **Comments**:
55.              {% for comment in steps.get_issue_comments.output %}- **{{ comment.user.login }}**: {{ comment.body }}
56.              {% endfor %}
57.              ---

59.              Provide a detailed, step-by-step solution in Markdown that includes:
60.              1. A brief summary of the problem
61.              2. Root cause analysis (if determinable)
62.              3. A concrete, actionable solution with code examples where applicable
63.              4. Any relevant references or best practices
64.            systemPrompt: |
65.              You are an expert software engineer and open-source contributor.
66.              Your task is to analyze GitHub issues and suggest concrete, actionable solutions.
67.              Always respond in well-structured Markdown.
68.        - name: set_output
69.          type: data.set
70.          with:
71.            issue_number: "{{ steps.extract_issue.output.number }}"
72.            issue_title: "{{ steps.extract_issue.output.title }}"
73.            issue_url: "{{ steps.extract_issue.output.html_url }}"
74.            issue_state: "{{ steps.extract_issue.output.state }}"
75.            suggested_solution: "{{ steps.analyze_issue.output.content }}"

`AI代码解释![](https://csdnimg.cn/release/blogv2/dist/pc/img/runCode/icon-arrowwhite.png)收起代码块![](https://csdnimg.cn/release/blogv2/dist/pc/img/arrowup-line-top-White.png)

上面创建的 GitHub Issue Analyzer1 workflow。

步骤三:创建一个 github tool

我们使用上面创建的 workflow 来创建一个叫做 github_issue_analyzer1 的 tool。这个 tool 将会被在下面的 agent 里所使用:

步骤四:创建一个 skill

  • Description:
`Analyze all GitHub issues (open and closed) from the liu-xiao-guo/elastic_multimodal_search repository. Fetches issue comments and generates AI-suggested solutions for each issue. Use when the user asks to analyze, review, or get solutions for repository issues.`AI代码解释
  • Instructions
``

1.  ## When to Use

3.  Use this skill when the user:
4.  - Asks to analyze, investigate, or get AI-suggested solutions for GitHub issues
5.  - Wants to review all open and/or closed issues in the `liu-xiao-guo/elastic_multimodal_search` repository
6.  - Asks "what issues are there?", "analyze the issues", or "run the issue analyzer"

8.  Do **not** use this skill for:
9.  - General GitHub questions unrelated to issue analysis
10.  - Creating or updating GitHub issues (not supported)
11.  - Analyzing a single specific issue by number (the workflow analyzes all issues at once)

13.  ## Available Tools

15.  - **platform.core.execute_workflow** — Runs the `github-issue-analyzer1` workflow with no inputs. It fetches all issues (open and closed), retrieves their comments, and returns an AI-generated solution for each.

17.  ## Workflow

19.  1. **No inputs required** — the workflow runs against the hardcoded repo `liu-xiao-guo/elastic_multimodal_search`.

21.  2. **Run the workflow** by calling `platform.core.execute_workflow` with:
22.     - `attachmentId`: `github-issue-analyzer1`
23.     - `inputs`: `{}` (no inputs needed)

25.  3. **Present the results** to the user. The workflow output is an array of results, one per issue. Each entry contains:
26.     - `issue_number` — the GitHub issue number
27.     - `issue_title` — the issue title
28.     - `issue_url` — clickable link to the issue on GitHub
29.     - `issue_state` — `open` or `closed`
30.     - `suggested_solution` — AI-generated Markdown solution (includes analysis of comments)

32.     Format the response clearly:
33.     - List each issue with its number, title, state, and URL
34.     - Present the suggested solution in full Markdown for each issue
35.     - Highlight whether comments were factored into the solution

37.  ## Example

39.  User: *"Analyze all the GitHub issues"*

41.  1. Call `platform.core.execute_workflow` with `attachmentId: github-issue-analyzer1`, `inputs: {}`.
42.  2. For each issue in the output, reply with:
43.     - **Issue #N** — [title](url) — `open` / `closed`
44.     - AI-suggested solution in Markdown

``AI代码解释![](https://csdnimg.cn/release/blogv2/dist/pc/img/runCode/icon-arrowwhite.png)

步骤五:创建 github demo 智能体

创建智能体

我们运用上面创建的 tool 及 skill 来创建一个 github demo 智能体:

添加 tool

添加 skill

这样我们就添加了  skill。

步骤六:使用 github demo 智能体

我们使用如下的步骤来使用上面的智能体:

我们来使用智能体来查看一下我们的 issues:

上面显示的是我们的 issue:Cannot upload a picture when searching for an image · Issue #1 · liu-xiao-guo/elastic_multimodal_search · GitHub

我们使用如下的方法来查看:

`Cannot upload a picture when searching for an image, what is the solution to it?`AI代码解释

`How many issues are there? what are the states for them?`AI代码解释

总结

我们可以在 Kibana 中轻松地配置 connnectors,并使它们很方便地引入到我们的 agents 里。这样我们使用连接器里的数据结合企业数据,更容易得到我们想要的结果。