Step-by-Step Guide to Building AI Assistants with Alibaba Cloud Elasticsearch Agent Builder

Agent Builder is Elastic's intelligent agent‑building capability that leverages large language models (LLM) and Elasticsearch data for retrieval, analysis, and interaction. It enables rapid creation of business‑scenario assistants such as operations Q&A, log analysis, security analysis, and knowledge‑base assistants.

Prerequisites

You need an Elasticsearch instance version 9.3.2 or later. If you have not created an instance, follow the Alibaba Cloud Elasticsearch instance creation guide.

1. Create an Agent in Kibana

Log in to the Alibaba Cloud Elasticsearch management console.

In the console, click the target Elasticsearch instance .

Navigate to Configuration > Visual Control and ensure the Kibana public‑access whitelist is configured (see the quick‑start guide if needed).

Click Modify Configuration , then log in to the Kibana page with your username and password.

Open the left‑hand menu, select Elasticsearch > Agents .

Click Connector > AI Connector and fill in the connector information as follows:

Connector name : any custom name.

Service type : select OpenAI . (The AI Search Open Platform’s large‑model interface is compatible with the OpenAI API definition.)

Model ID : enter the model name, e.g., qwen-turbo, qwen-plus, or qwen3-max. Note that ops-qwen-turbo is not supported in AI assistants.

URL :

http://{model_service_address}/compatible-mode/v1/chat/completions

. Obtain the model service address from the AI Service Center > Model Management page.

API Key : copy the string shown as “Model Call API Key” in the AI Service Center > Model Management page.

Additional settings : set Task type to chat_completion .

After filling the fields, click Save and Test . The page will display three steps: Create Operation , Run Test , and Result . When the result area shows “Test execution succeeded” and a message “Created ‘ your‑connector‑name ’”, the connector is successfully created.

Return to Elasticsearch > Agents , refresh the page, and you will see the dialogue interface “How can I help you?” with the newly connected connector and Elastic AI Agent, indicating the agent is ready.

2. Usage Example

Once the agent is created, you can interact with Elasticsearch data using natural language. For instance, if an index api_access_logs stores historical API request logs, you can type:

"Analyze the request history and identify the slowest API endpoint"

The agent performs the following reasoning process:

Calls the built‑in tool platform.core.search to discover relevant data sources.

Locates the target index api_access_logs.

Automatically generates an ES|QL query that groups logs by API name and computes average response time, maximum response time, and request count.

The agent returns the analysis results, including:

Details of the slowest request (API name, path, latency, status code, etc.).

Performance statistics for each API (average latency, max latency, request count) visualized as a bar chart.

Optimization suggestions for investigating slow‑query APIs.

Conclusion

Alibaba Cloud Elasticsearch combines innovations such as BBQ quantification, the FalconSeek engine, and the Retrievers framework to deliver accurate, affordable, and manageable enterprise‑level AI search. For technical leaders building next‑generation intelligent applications, Alibaba Cloud Elasticsearch offers an optimal balance of performance, cost, and security.

For more information, refer to the official documentation links provided in the original article.