# opensearch-text

Part of **OPENSEARCH**

# OpenSearch Model and AI Services Console Guide

## Operations Overview

| Operation | Console Navigation Path | Prerequisites | Description |
|-----------|------------------------|---------------|-------------|
| Activate AI Search Service and Create Workspace | Console > AI Search Open Platform > Service Plaza | Valid Alibaba Cloud account with appropriate permissions | Activates the AI Search Open Platform and creates a workspace for data isolation |
| Deploy Model Service | Console > AI Search Open Platform > Model Service > Service Deployment | AI Search Open Platform account; RAM users need Model Service-Service Deployment permission | Deploys hosted inference services for text vectorization, reranking, and multimodal vector models |
| Customize Embedding Model | Console > AI Search > Model Service > Create Model | RAM user with model service permissions; MaxCompute project or OSS Bucket with training data; AccessKey credentials | Trains custom text embedding or dimensionality reduction models using business data |
| Configure NL2SQL Service | AI Search Open Platform > Instance Info > RAG Model Service Configuration > Create New Service | Access to console; understanding of database schema | Sets up natural language to SQL service with table configuration, training samples, and custom rules |
| Create Evaluation Task | Console > AI Search Open Platform > Effect Evaluation > Create Evaluation Task | AI Search Open Platform service activated | Creates tasks to evaluate RAG pipeline performance using provided datasets |
| Browse and Test Services in Experience Center | Console > Open Platform for AI Search > Experience Center | None (no login required) | Allows testing of document parsing, multimodal vector, and other AI services without authentication |
| Enable Agentic Search | Console > AI Search Open Platform > Agentic Search | Access to console; valid region selected | Configures and runs context-aware search tasks with different modes and knowledge bases |
| Implement RAG Pipeline | Console > OpenSearch > Instances > Create Instance | Ability to purchase OpenSearch Recall and Conversational Search instances | Sets up Retrieval-Augmented Generation pipeline with proper instance configuration |
| Create Enterprise Chatbot | Console > OpenSearch > LLM-Based Conversational Search > Create Chatbot | Data imported into Conversational Search; DingTalk/Lark admin access | Creates chatbots for enterprise group chats in DingTalk or Lark |

## Step-by-Step Instructions

### Activate AI Search Service and Create Workspace

**Navigation**: Console > AI Search Open Platform > Service Plaza

**Prerequisites**:
- Valid Alibaba Cloud account with appropriate permissions

1. Log in to the AI Search Open Platform console
   - Element: **AI Search Open Platform** (link) — top navigation bar

2. Click the activation button in the Activation Reminder banner
   - Element: **Activate** (button) — top of Service Plaza page

3. Read the Search Development Service Platform Agreement and sign it
   - Element: **Search Development Service Platform Agreement** (link) — in the agreement signing dialog box

4. Confirm the agreement by clicking the Activate button
   - Element: **Activate** (button) — agreement signing dialog box

5. Navigate to the Service Plaza page and click the workspace management button
   - Element: **Manage Workspaces** (button) — upper-right corner of Service Plaza page

6. Click the Create Workspace button and enter a workspace name
   - Element: **Create Workspace** (button) — workspace management panel
   - Notes: The system automatically creates a Default workspace after first-time activation.

7. Confirm the creation by clicking the Confirm button
   - Element: **Confirm** (button) — create workspace dialog
   - Notes: After creation, you can assign RAM user permissions for enhanced security.

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| Workspace Name | text_input | Yes | — | The name of the new workspace to be created. Must be unique within the account. |

### Deploy Model Service

**Navigation**: Console > AI Search Open Platform > Model Service > Service Deployment

**Prerequisites**:
- An AI Search Open Platform account
- (RAM users only) The Model Service-Service Deployment permission granted to your RAM user

1. Go to Model Service > Service Deployment and click Deploy Service
   - Element: **Deploy Service** (button) — top-right corner of the Service Deployment page

2. Configure service settings including name, region, resource type, and view estimated price
   - Element: **Service name** (text_input) — main content area
   - Notes: Only Germany (Frankfurt) is currently supported as the deployment region.

3. Click Deploy to start provisioning the service
   - Element: **Deploy** (button) — bottom of the Deploy Service form

4. Click Manage from the service list to view service details
   - Element: **Manage** (button) — service status row in the service list
   - Notes: Available actions depend on the current service status (e.g., Normal, Deploying, Deployment Failed)

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| Service name | text | Yes | — | A name to identify this deployment |
| Deployment region | dropdown | Yes | Germany (Frankfurt) | The region where the service runs. Currently, only Germany (Frankfurt) is supported. |
| Resource type | dropdown | Yes | — | The compute resource type for model inference |
| Estimated price | text | No | — | The estimated cost for the selected configuration |

### Customize Embedding Model

**Navigation**: Console > AI Search > Model Service > Create Model

**Prerequisites**:
- RAM user with required model service permissions
- MaxCompute project or OSS Bucket with training data
- AccessKey ID and secret with read/write permissions

1. Navigate to the AI Search Open Platform console and go to Model Service section
   - Element: **AI Search Open Platform** (link) — top navigation bar

2. Click on the 'Create Model' button
   - Element: **Create Model** (button) — main content area

3. Select model type: Embedding Dimensionality Reduction or Text Embedding
   - Element: **Embedding Dimensionality Reduction** (radio) — form fields

4. Configure data source (MaxCompute or OSS) and enter required credentials
   - Element: **Training data source** (dropdown) — form fields
   - Notes: For MaxCompute: provide AccessKey ID, AccessKey secret, project name, table name, and partition. For OSS: provide region, bucket name, and data path.

5. Confirm creation and start training
   - Element: **Confirm** (button) — confirmation dialog
   - Notes: Click 'Create and Train' to start immediate training, or 'Create' to schedule later.

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| Model name | text | Yes | — | The name used to invoke the embedding dimensionality reduction service. |
| Model type | radio | Yes | Embedding Dimensionality Reduction, Text Embedding | The type of model to train. |
| Base model | dropdown | Yes | ops-embedding-dim-reduction-001, ops-text-embedding-001 | The base model for training, such as ops-embedding-dim-reduction-001. |
| Training data source | dropdown | Yes | MaxCompute, OSS | Specifies where the training data is stored. |
| Region | dropdown | Yes | — | The region where your MaxCompute project or OSS Bucket is located. |
| Project name | text | Yes | — | The name of your project in MaxCompute. |
| AccessKey ID | text | Yes | — | The AccessKey ID of the Alibaba Cloud account or RAM user with read and write permissions for MaxCompute. |
| AccessKey secret | text | Yes | — | The AccessKey secret that corresponds to the AccessKey ID. |
| Table name | text | Yes | — | The name of the table in MaxCompute that stores your training data. |
| Table partition | text | No | — | The partition information of the table. |
| Training fields | text | Yes | — | To select the primary key field and String-type vector fields, grant GetTableFields permission to the RAM user. |
| Query-doc pairs | text | Yes | — | For the required data format, see the sample data in the console. |
| OSS Bucket | text | Yes | — | The name of your OSS Bucket. |
| Doc data | text | Yes | — | The data in OSS used for training. |
| OSS Endpoint | text | No | — | This value is automatically generated after you configure the preceding parameters. |

### Configure NL2SQL Service

**Navigation**: AI Search Open Platform > Instance Info > RAG Model Service Configuration > Create New Service

**Prerequisites**:
- Access to the AI Search Open Platform console
- Basic understanding of database schema and business data structure

1. Navigate to the RAG Model Service Configuration page and click Create New Service
   - Element: **Create New Service** (button) — top-right corner

2. Enter a service name such as student_info_analysis
   - Element: **Service Name** (text_input) — main content area
   - Notes: The name must be unique and follow naming conventions: lowercase letters, digits, underscores only, max 30 characters

3. Click Next to proceed to the next step
   - Element: **Next** (button) — bottom of form

4. Configure basic table information using JSON format
   - Element: **Basic Table Configuration** (text_input) — main content area
   - Notes: Must include table name, columns with type, description, example, and value_mapping. Table and column names must start with lowercase letter and contain only lowercase letters, digits, or underscores.

5. Configure table join relationships using JSON array
   - Element: **Table Join Configuration** (text_input) — main content area
   - Notes: Format: ["table1.field=table2.field"]

6. Click Next to proceed to custom rules and training samples
   - Element: **Next** (button) — bottom of form

7. Add training samples with queries and corresponding SQL statements
   - Element: **Training Sample Configuration** (text_input) — main content area
   - Notes: Use JSON array format with 'query' and 'sql' fields

8. Add custom rules (glossary) for business-specific terms and concepts
   - Element: **Custom Rule Configuration** (text_input) — main content area
   - Notes: Use JSON array with 'key' and 'value' fields; e.g., "Model student": "students.id <= 10"

9. Click Finish to complete configuration and activate the service
   - Element: **Finish** (button) — bottom of form

10. Click Activate to start activation process
    - Element: **Activate** (button) — service list
    - Notes: Status changes from Activating to Activated after validation

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| Service Name | text | Yes | — | Unique identifier for the NL2SQL service. Must be lowercase, alphanumeric, or underscore, max 30 characters. |
| Basic Table Configuration | text | Yes | — | JSON-formatted configuration of tables, including table name, columns with type, description, example, and value mapping. |
| Table Join Configuration | text | No | — | JSON array specifying join conditions between tables, e.g., ["students.class=schools.class"] |
| Training Sample Configuration | text | No | — | JSON array of query-SQL pairs to improve model accuracy. Each object contains 'query' and 'sql' fields. |
| Custom Rule Configuration | text | No | — | JSON array of business-specific term definitions. Each object has 'key' (term) and 'value' (definition). |

### Create Evaluation Task

**Navigation**: Console > AI Search Open Platform > Effect Evaluation > Create Evaluation Task

**Prerequisites**:
- AI Search Open Platform service must be activated

1. Log on to the AI Search Open Platform console
   - Element: **AI Search Open Platform console** (link) — top navigation

2. Select the China (Shanghai) region and navigate to the RAG product section
   - Element: **China (Shanghai)** (dropdown) — top-right corner
   - Notes: AI Search Open Platform is currently available only in China (Shanghai) and Germany (Frankfurt). Users in other Chinese regions can access via VPC.

3. Select the target workspace
   - Element: **workspace** (dropdown) — main content area
   - Notes: The system automatically creates a Default workspace after first activation. Additional workspaces can be created.

4. In the left-side navigation, choose Effect Evaluation and click Create Evaluation Task
   - Element: **Effect Evaluation** (menu) — left-side navigation panel

5. On the Create Evaluation Task page, enter a task name and upload an evaluation dataset in the specified format
   - Element: **Create Evaluation Task** (button) — main content area
   - Notes: Dataset must follow the provided template. Maximum of 200 valid records. Reference answer is optional but all questions in a dataset must be consistent (all have or none have a reference answer).

6. Click Confirm to create the evaluation task
   - Element: **Confirm** (button) — bottom of form

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| Task Name | text | Yes | — | A unique name for the evaluation task. |
| Evaluation Dataset | file | Yes | — | Upload a dataset file in the exact format of the sample template. Must contain 'question', 'standard_answer' (optional), 'recall_docs', and 'model_answer' fields. |

### Browse and Test Services in Experience Center

**Navigation**: Console > Open Platform for AI Search > Experience Center

**Prerequisites**:
- No login required to access the Experience Center
- File size must not exceed 20 MB
- Supported file formats: TXT, PDF, HTML, DOC, DOCX, PPT, PPTX

1. Log on to the Open Platform for AI Search console
   - Element: **Open Platform for AI Search console** (link) — top navigation

2. Select Experience Center from the left-side navigation panel
   - Element: **Experience Center** (menu) — left-side navigation panel

3. Select a service category and specific service under Server ID
   - Element: **document-analyze** (dropdown) — main content area
   - Notes: For document parsing, select 'document-analyze'. For multimodal vector, select 'multi-modal-embedding'.

4. Upload a local file or provide a URL for testing
   - Element: **Upload a local file** (button) — main content area
   - Notes: Files are automatically deleted after seven days. Use web link import only as permitted by applicable laws and regulations.

5. Click Get Result to trigger the service call
   - Element: **Get Result** (button) — main content area

6. View the response data or download the API code
   - Element: **Response data** (link) — result section
   - Notes: Use Copy or Download to save the response or code locally.

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| Server type | dropdown | Yes | document-analyze, multi-modal-embedding | Selects the category of service to test. |
| Server ID | dropdown | Yes | M2-Encoder-multimodal vector model, M2-Encoder-Large-multimodal vector model, GME multimodal vector-Qwen2-VL-2B, Multimodal vector-ops-mm-embedding-v1-2b, Multimodal vector-ops-mm-embedding-v1-7b, E-commerce multimodal vector-ops-mm-embedding-ecom-001, Face multimodal vector-ops-mm-embedding-face-001, Document parsing service 001, Document parsing service 002 | Selects the specific service instance to use. |
| Input type | radio | Yes | Text, Image, Multiple images with text | Specifies the input format for the multimodal vector service. |

### Enable Agentic Search

**Navigation**: Console > AI Search Open Platform > Agentic Search

**Prerequisites**:
- Access to the AI Search Open Platform console
- A valid region selected for the service

1. Log in to the AI Search Open Platform and click the 'Open' button in the 'Use Notice' section
   - Element: **Open** (button) — Use Notice section

2. Select a region from the dropdown and click the 'Open' button
   - Element: **Open** (button) — Region selection area

3. In the left navigation pane, click on 'Agentic Search'
   - Element: **Agentic Search** (menu) — left navigation pane

4. In the demo case section, click the 'Try it' button
   - Element: **Try it** (button) — demo case section

5. Click the upload icon to upload a local file or enter an online document URL
   - Element: **upload icon** (button) — Upload a file section
   - Notes: Image shows a paperclip icon

6. Select a mode by clicking on one of the mode icons: planning, conversational, or adaptive
   - Element: **planning mode icon** (radio) — Select a mode section
   - Notes: Image shows a checklist icon

7. Click the output format icon to set the output format to Markdown or HTML
   - Element: **output format icon** (button) — Set the output format section
   - Notes: Image shows a document icon

8. Select an existing knowledge base from the dropdown
   - Element: **knowledge base dropdown** (dropdown) — Select a knowledge base section
   - Notes: Link provided to create a new knowledge base

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| Region | dropdown | Yes | — | Choose the region where the Agentic Search service will be activated |
| Mode | radio | Yes | planning mode, conversational mode, adaptive mode | Select the execution mode for the task |
| Output Format | dropdown | No | Markdown, HTML | Set the format of the generated output |
| Knowledge Base | dropdown | Yes | — | Choose an existing knowledge base to use for the task |

### Implement RAG Pipeline

**Navigation**: Console > OpenSearch > Instances > Create Instance

**Prerequisites**:
- An OpenSearch Recall Engine Edition instance or the ability to purchase one
- An OpenSearch Conversational Search Edition instance purchased separately
- AccessKey ID and Secret for both instances
- Basic understanding of document processing and vector search concepts

1. Purchase a new OpenSearch Recall Engine Edition instance
   - Element: **Purchase an OpenSearch Recall Engine Edition instance** (link) — Instance List page

2. Click Configure on the Instance List page
   - Element: **Configure** (button) — Instance List page

3. Set table name, number of shards, and number of data update resources
   - Element: **table name** (text_input) — Basic table information section
   - Notes: You receive two data update resources free of charge; additional ones are billed.

4. Configure data source as API data source
   - Element: **API data source** (dropdown) — Data synchronization section

5. Proceed to index structure configuration
   - Element: **Next** (button) — Bottom of the data source configuration page

6. Define required fields: doc_id, split_content, split_content_embedding
   - Element: **Index structure configuration** (text_input) — Index structure configuration page
   - Notes: Ensure split_content_embedding is a multi-value FLOAT vector field with comma-separated values.

7. Set index types: PRIMARYKEY64 for doc_id, PACK/TEXT for content fields, CUSTOMIZED for split_content_embedding
   - Element: **CUSTOMIZED index** (dropdown) — Index configuration section
   - Notes: Set vector index dimension to 1536 for split_content_embedding.

8. Click Confirm and Create to finalize the instance
   - Element: **Confirm and Create** (button) — Bottom of the configuration page

9. Check creation progress on the Feature Extensions > Change History page
   - Element: **Feature Extensions > Change History** (menu) — Top navigation bar

10. Purchase an OpenSearch Conversational Search Edition instance
    - Element: **Build an enterprise knowledge base Q&A system in the console** (link) — Documentation section
    - Notes: No special configuration needed after purchase.

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| table name | text | Yes | — | Customize the name of the data table. |
| number of shards | number | Yes | — | Set the number of shards for the table. |
| number of data update resources | number | Yes | — | Number of resources used for data updates; first two are free. |
| data source type | dropdown | Yes | Full data source, API data source | Select the method for data ingestion. |
| index type for doc_id | dropdown | Yes | PRIMARYKEY64 | Must be set to PRIMARYKEY64 for the primary key field. |
| index type for content fields | dropdown | Yes | PACK, TEXT | Choose index type for text fields like content and split_content. |
| index type for split_content_embedding | dropdown | Yes | CUSTOMIZED | Must be set to CUSTOMIZED for vector fields. |
| vector index dimension | number | Yes | — | Set the dimension of the vector index for embedding fields. |

### Create Enterprise Chatbot

**Navigation**: Console > OpenSearch > LLM-Based Conversational Search > Create Chatbot

**Prerequisites**:
- Data imported into OpenSearch LLM-Based Conversational Search Edition
- Access to DingTalk or Lark workspace with admin privileges
- Enterprise account with sufficient permissions for bot creation

1. Navigate to the OpenSearch console and select the LLM-Based Conversational Search service.
   - Element: **OpenSearch** (menu) — left navigation panel

2. Click on 'Create Chatbot' in the LLM-Based Conversational Search section.
   - Element: **Create Chatbot** (button) — main content area
   - Notes: The button is visible only after data import is complete.

3. Select the target platform: DingTalk or Lark.
   - Element: **Platform** (dropdown) — form fields
   - Notes: Users must choose one platform at a time.

4. Follow the guided wizard to configure the chatbot name, description, and associated knowledge base.
   - Element: **Next** (button) — bottom of form
   - Notes: The wizard includes visual previews of the final chatbot behavior.

| Parameter | Type | Required | Options/Values | Description |
|-----------|------|----------|----------------|-------------|
| Chatbot Name | text_input | Yes | — | The name displayed for the chatbot in group chats. |
| Description | text_input | No | — | A brief description of the chatbot's purpose. |
| Knowledge Base | dropdown | Yes | Internal Documents, Marketing Materials, Customer Support FAQ, Product Catalog | Select the data source that the chatbot will use for responses. |

## FAQ

Q: Where can I find the Service ID for my deployed model?
A: After deploying a model service, go to the Service Deployment page and click "Manage" on your service row. The Service ID is displayed in the service details page.

Q: Can I modify the configuration of a deployed model service?
A: Most configuration parameters cannot be modified after deployment. You would need to create a new deployment with the desired settings.

Q: What happens if I leave the Knowledge Base field empty when enabling Agentic Search?
A: The Knowledge Base field is required. You must select an existing knowledge base or create a new one before proceeding with the Agentic Search task.

Q: Do I need to activate the AI Search Open Platform before using the Experience Center?
A: No, the Experience Center can be accessed without login or service activation for trial purposes. However, production usage requires service activation.

Q: What permissions does a RAM user need to deploy model services?
A: RAM users need the "Model Service-Service Deployment" permission to deploy model services in the AI Search Open Platform.

## Pricing & Billing

### Billing Model
Services are billed based on actual usage with different pricing models:
- Per request: Document splitting, NL2SQL, Agentic Search, RAG pipeline components
- Per CU (Computing Unit): Model customization training
- Per instance hour: Retrieval Engine instances

### Price Reference

| Service | Input Price | Output Price |
|---------|-------------|--------------|
| Document splitting | 0.0005 / | — |
| NL2SQL service | 0.0001 / | 0.0001 / |
| Agentic Search | 0.002 /tokens | 0.004 /tokens |
| Text vectorization | 0.002 /tokens | 0.002 /tokens |
| Reranking | 0.005 /tokens | 0.005 /tokens |
| Multimodal vector | 0.01 /tokens | 0.01 /tokens |
| OpenSearch Recall Engine Edition | 0.002 / | 0.003 / |
| OpenSearch Conversational Search Edition | 0.005 / | 0.010 / |
| Enterprise chatbot | 0.001 / | 0.002 / |
| Model customization | CNY 3.87 per CU | — |
| Retrieval Engine Standard | 0.08 / | — |
| Retrieval Engine High-Performance | 0.25 / | — |

### Free Tier
- Document splitting: 10,000 free requests per month
- NL2SQL service: 1,000 free requests per month
- Agentic Search: 1 million tokens free per month
- RAG pipeline components: 1 million tokens free per month
- OpenSearch instances: 1,000 free requests per month
- Enterprise chatbot: 500 free requests per month
- Experience Center: Free trial access without login

### Billing Notes
- Minimum charge per request is 100 tokens for token-based services
- Async tasks are billed upon completion
- Retrieval Engine instances have a minimum 1-hour charge
- Model customization costs depend on training data volume and dimensionality
- Web link imports in Experience Center must comply with applicable laws and regulations