You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This is an experimental UI feature. For updates on the progress of the feature or if you want to leave feedback, join the discussion on the [OpenSearch forum](https://forum.opensearch.org/).
13
+
This is an experimental UI feature. For updates on the progress of the feature or if you want to leave feedback, join the discussion on the [OpenSearch forum](https://forum.opensearch.org/).
14
14
{: .warning}
15
15
16
-
[Agentic search]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/) lets you ask questions in natural language and have OpenSearch agents plan and execute the retrieval automatically. OpenSearch Dashboards offers an intuitive UI for configuring agents, equipping agents with different tools, and executing agentic searches.
16
+
[Agentic search]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/) allows you to ask questions in natural language and have OpenSearch agents plan and execute retrieval automatically. This feature combines large language models (LLMs) with OpenSearch's search capabilities to provide intelligent, context-aware search experiences.
17
+
18
+
OpenSearch Dashboards provides an intuitive interface for configuring agents, equipping agents with different tools, executing agentic searches, and integrating agentic search into your applications.
19
+
20
+
The following image shows the agentic search workflow interface with agent configuration options on the left and search execution capabilities on the right.
Before configuring agentic search, ensure that you fulfill the following prerequisites.
21
27
22
28
### Provision ML resources
23
29
24
-
To configure new agents, be sure to first provision appropriate models. For working examples, see [Model configuration]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/agent-customization/#model-configuration).
30
+
To configure new agents, first provision appropriate models. For working examples, see [Model configuration]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/agent-customization/#model-configuration).
25
31
26
32
### Ingest data
27
33
28
-
Ensure that you have a sufficient number of documents in your cluster to reasonably evaluate your agentic searches. For more information, see [Agentic search]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/).
34
+
Ensure that you have a sufficient number of documents in your cluster to reasonably evaluate your agentic searches.
35
+
36
+
## Accessing the plugin
37
+
38
+
To access the plugin, go to **OpenSearch Dashboards** and select **OpenSearch Plugins** > **AI Search Flows** from the top menu.
39
+
40
+
## Configuring agents
41
+
42
+
Agents are highly customizable and can be configured in multiple ways depending on your use case. The following image shows a fully configured conversational agent.
[Flow agents]({{site.url}}{{site.baseurl}}/ml-commons-plugin/agents-tools/agents/flow/) are optimized for speed and simplicity in query generation. [Conversational agents]({{site.url}}{{site.baseurl}}/ml-commons-plugin/agents-tools/agents/conversational/) can be configured with more tools and functionality for deeper reasoning. Agent types can only be configured **once** during agent creation.
49
+
50
+
When configuring flow agents, some models require you to manually add a suitable `response_filter` in the **Query Planning** tool. For more information, see [Register a flow agent]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/flow-agent/#step-4-register-a-flow-agent). Supported providers are OpenAI and Amazon Bedrock Converse.
51
+
{: .note}
52
+
53
+
### Models
54
+
55
+
In conversational agents, models are responsible for intelligent reasoning, including tool orchestration, connecting to external sources, and generating appropriate query domain-specific language (DSL) queries. In flow agents, tools run sequentially, and models in the **Query Planning** tool are used only for query generation. Different models are optimized for different scenarios: cost-efficient, fast inference versus resource-intensive, deep-reasoning approaches.
56
+
57
+
For a list of suggested models compatible with agentic search, see [Model configuration]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/agent-customization/#model-configuration).
58
+
59
+
### Tools
60
+
61
+
All agents require the **Query Planning** tool to execute agentic searches. Query generation can be entirely LLM-generated (default) or guided by predefined search templates. Search templates help maintain control over generated queries by directing the model to use known, tested, and performant query patterns.
62
+
63
+
The following image shows the search templates configuration interface within the **Query Planning** tool.
Other prebuilt tools are available for conversational agents, including **Search Index**, **List Index**, **Index Mapping**, and **Web Search**. Enable these to extend your agent's capabilities. While these represent the most commonly used tools in agentic search, several additional [OpenSearch tools are available]({{site.url}}{{site.baseurl}}/ml-commons-plugin/agents-tools/tools/index/) and can be configured manually in the **JSON** view.
68
+
69
+
### MCP servers
70
+
71
+
To allow conversational agents to access more external tools, integrate them with Model Context Protocol (MCP) servers. To limit the tools that the agents can access, configure filters for each server under **Tool filters**. For more information, see [Using external MCP servers]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/mcp-server).
72
+
73
+
## Running agentic searches
74
+
75
+
Test how agents perform with different indexes and search queries. Collapse the **Configure agent** panel to focus on executing searches and analyzing results. The following image shows a search for `mens blue shirts` against the `demo_amazon_fashion` index, including relevant result images.
Try different indexes in your cluster. To view index details, select the **Inspect** button. For conversational agents, you can select **All indexes** and let the agent choose the appropriate index.
82
+
83
+
### Agent
84
+
85
+
Try different agents you've created. To view agent details, select the **Inspect** button.
86
+
87
+
### Query
88
+
89
+
Test different natural language queries. To specify fields for the agent to focus on in your selected index, select **Add query fields**. To edit the full `agentic` search query directly, toggle to the **JSON** view. For conversational agents, select **Continue conversation** after a search to maintain context for future searches. To discard conversational history and start again, select **Clear conversation**.
90
+
91
+
### Running searches
92
+
93
+
To run an agentic search, select **Search**. The process may take several seconds while the agent reasons about the query, analyzes index mappings, and performs tool orchestration. If the search takes too long or you want to try a different search, select **Stop**.
94
+
95
+
After the search completes, view the results in the following sections:
96
+
97
+
- **Generated query**: The query DSL that the agent generated and executed against your cluster
98
+
- **Search results**: The search response with available tabs based on the results:
99
+
- **Aggregations**: Displayed when the response contains aggregations
100
+
- **Visual hits**: Displayed when document hits contain images
101
+
- **Hits**: Displayed when the response contains any hits
102
+
- **Raw response**: Always available for detailed inspection
103
+
104
+
For conversational agents, select **View agent summary** to see a step-by-step breakdown of the agent's actions, including the sequence of tools used and the reasoning behind each step.
105
+
106
+
The following image shows the search results interface with different viewing tabs and the agent summary option.
Select **Export** to view all of the underlying resources needed for using agentic search in your downstream application, including the agent and search pipeline details.
113
+
114
+
The following image shows the export dialog with code examples for integrating agentic search into your applications.
This example uses a deployed OpenAI GPT-5 model described in [this documentation]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/agent-customization/#gpt-5-recommended).
120
+
This example uses a deployed OpenAI [GPT-5 model]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/agent-customization/#gpt-5-recommended) and an index generated from the [fashion product images dataset](https://www.kaggle.com/datasets/paramaggarwal/fashion-product-images-dataset).
33
121
{: .note}
34
122
35
-
To build and test your agentic search workflowin OpenSearch Dashboards, follow these steps:
123
+
1. Navigate to the **AI Search Flows** plugin. On the **Workflows** page, select the **New workflow** tab, as shown in the following image. In the **Agentic Search** template, select **Create**.
36
124
37
-
1. Go to **OpenSearch Dashboards** and select **OpenSearch Plugins** > **AI Search Flows** from the top menu.
38
-
1. On the **Workflows** page, select the **New workflow** tab, as shown in the following image. In the **Agentic Search** template, select **Create**.
1. Provide a unique workflow **Name** and an optional **Description**, as shown in the following image. Then select **Create** to create your workflow. You are automatically directed to the workflow editor, where you can begin configuring the agent.
1. Under **Configure agent**, select **Create new agent**, as shown in the following image. Once you configure an agent, you can select or update existing agents. For full agent customization, toggle to the `JSON` view and edit directly.
1. Under **Agent**, enter a unique **Name** and **Description** for the agent, as shown in the following image. Under **Tools** > **Query Planning** > **Query planning model**, select **OpenAI GPT-5**. Then select **Create agent**.
1. Under **Test flow** > **Index**, select the index you'd like to search, as shown in the following image. Under **Test flow** > **Query**, enter a natural language query. Optionally, specify query fields for the agent to search in your selected index. To edit the full `agentic` search query directly, toggle to the `JSON` view. Then select **Search**. The agent may take several seconds to execute.
2. Provide a unique workflow **Name** and an optional **Description**, as shown in the following image. Then select **Create** to create your workflow. You are automatically directed to the workflow editor, where you can begin configuring the agent.
3. Under **Configure agent**, select **Create new agent**. Enter a unique **Name**, optionally provide a **Description**, and under **Model**, select the deployed **OpenAI GPT-5** model. Finally, select **Create agent** at the bottom.
4. Under **Agentic search**, select the index you'd like to search against, or leave the default as **All indices** and let the agent decide. The agent will already be selected.
133
+
5. Under **Query**, enter a natural language query about your data, as shown in the following image. Select **Search** to run an agentic search.
7. Optionally, tune your agent by trying different tools, models, and MCP server integrations to see how the agent performs on different queries against your data.
151
+
152
+
## Next steps
153
+
154
+
- Try agentic search on the [ML Playground](https://ml.playground.opensearch.org/app/opensearch-flow#/workflows/WAmxgJoBWVNV3bhKKnGx?configureAgent=false).
155
+
- Learn more about [agentic search]({{site.url}}{{site.baseurl}}/vector-search/ai-search/agentic-search/).
156
+
- Read [this blog post on agentic search](https://opensearch.org/blog/introducing-agentic-search-in-opensearch-transforming-data-interaction-through-natural-language/).
0 commit comments