OSS Vector Search & AI Content Awareness

Leverage multimodal AI models to extract semantic descriptions and concise summaries from images, videos, audio, and documents stored in OSS Buckets. Build searchable vector indexes to enable advanced retrieval capabilities such as text-to-image and text-to-video search.

Prerequisites

Aliyun CLI (>= 3.3.1)

Pre-check: Aliyun CLI >= 3.3.1 required This skill uses Aliyun CLI for all OSS operations except opening MetaQuery. If Aliyun CLI is already installed, ossutil does not need to be installed separately. Run the following command to verify the version: aliyun version If not installed or the version is below 3.3.1, refer to references/cli-installation-guide.md for installation instructions, then run:

aliyun configure set --auto-plugin-install true

This configuration enables automatic plugin installation.

Update the aliyun ossutil plugin

Troubleshooting: --user-agent unknown flag error If you encounter Error: unknown flag: --user-agent when running a command, the ossutil plugin version is too old. Run the following command to update to the latest version:
aliyun ossutil update
After the update, re-run the original command.

# Python SDK dependencies
pip install oss2==2.19.1 alibabacloud-credentials==1.0.8

Notes:

Required: Aliyun CLI, Python dependencies oss2 and alibabacloud-credentials

Not required: ossutil

The only operation that requires Python: opening MetaQuery (AI Content Awareness + semantic search) Security Rules:

NEVER read, echo, or print AK/SK values (e.g., echo $ALIBABA_CLOUD_ACCESS_KEY_ID is FORBIDDEN)

NEVER ask the user to input AK/SK directly in the conversation or command line

Architecture

User Request -> OSS Bucket -> AI Content Awareness Engine -> Semantic Feature Extraction -> Vector Index -> Semantic Search
                    |
         Images/Videos/Audio/Docs -> Detailed Description (~100 chars) + Concise Summary (<=20 chars)

Core Components: OSS Bucket + Data Index + Vector Search + AI Content Awareness

Usage Restrictions

Supported Regions

Region Category	Region List
East China	cn-hangzhou, cn-shanghai
North China	cn-qingdao, cn-beijing, cn-zhangjiakou
South China	cn-shenzhen, cn-guangzhou
Southwest China	cn-chengdu
Other	cn-hongkong, ap-southeast-1 (Singapore), us-east-1 (Virginia)

Note: If the user's Bucket is in a region not listed above, vector-mode MetaQuery and content awareness cannot be enabled, and an EC Code 0037-00000001 error will be returned. Guide the user to create a new Bucket in a supported region.

File Types

Supported: Images, videos, audio, documents
Multipart uploads: Only objects that have been assembled via CompleteMultipartUpload are shown

Performance Reference

OSS Internal Bandwidth and QPS

Region	Internal Bandwidth	Default QPS
cn-beijing, cn-hangzhou, cn-shanghai, cn-shenzhen	10Gbps	1250
Other regions	1Gbps	1250

This bandwidth and QPS is provided exclusively for vector search and does not consume the Bucket's QoS quota.

Existing File Index Build Time

File Type	10 Million Files	100 Million Files	1 Billion Files
Structured data & images	2-3 hours	1 day	~10 days
Videos, documents, audio	2-3 days	7-9 days	-

Incremental Updates and Search Latency

Incremental updates: When QPS < 1250, latency is typically minutes to hours
Search response: Sub-second, default timeout 30 seconds

Dangerous Operation Confirmation

Before executing any of the following dangerous operations, you MUST confirm with the user first and obtain explicit consent before proceeding:

Delete Bucket: aliyun ossutil rm oss://<bucket-name> -b --user-agent AlibabaCloud-Agent-Skills -- Deletes the entire Bucket, irreversible
Delete Object: aliyun ossutil rm oss://<bucket-name>/<object-key> --user-agent AlibabaCloud-Agent-Skills -- Deletes a specific file
Batch Delete Objects: aliyun ossutil rm oss://<bucket-name>/ --recursive --user-agent AlibabaCloud-Agent-Skills -- Recursively deletes all files in the Bucket
Close MetaQuery: aliyun ossutil api close-meta-query --bucket <bucket-name> --user-agent AlibabaCloud-Agent-Skills -- Closes the metadata index; all indexed data will be cleared
Open MetaQuery: python scripts/open_metaquery.py --region <your-region> --bucket <your-bucket-name> --endpoint <your-endpoint> -- Opens the metadata index; existing data will start being indexed. If the bucket has more than 1000 objects, confirm with the user first.
Create Bucket: aliyun ossutil api put-bucket --bucket <bucket> --region <region-id> --user-agent AlibabaCloud-Agent-Skills -- Creates a Bucket

When confirming, explain the following to the user:

The specific operation to be performed
The scope of impact (which files/resources will be deleted or closed)
Whether the operation is reversible (most delete operations are irreversible)

RAM Permissions

See references/ram-policies.md

Critical Rules (Must Follow)

Rule 1: Opening MetaQuery MUST use the Python script

PROHIBITED:

aliyun ossutil api open-meta-query oss://my-bucket --mode semantic

REQUIRED:

python scripts/open_metaquery.py --region cn-hangzhou --bucket my-bucket

Reason: Only the Python script or SDK can correctly configure WorkflowParameters to enable AI Content Awareness (ImageInsightEnable and VideoInsightEnable). Without this, semantic search quality will be severely degraded.

Rule 2: Must ask the user when Bucket name conflicts

When creating a Bucket and encountering a BucketAlreadyExists error:

Immediately stop all subsequent operations
Inform the user: "The Bucket name is already taken"
Ask the user to choose:
- Option 1: Use the existing bucket (requires explicit user confirmation)
- Option 2: Choose a new bucket name (user provides the new name)
Wait for the user's response before continuing PROHIBITED:

Automatically modifying the bucket name (e.g., appending -2, -new, etc.)
Using an existing bucket without asking the user

Rule 3: Use Aliyun CLI by default for all operations except opening MetaQuery

The following operations should use Aliyun CLI by default:

Create Bucket
Query Bucket info
Query Bucket statistics
Upload files
Query MetaQuery status
Execute semantic search
Close MetaQuery
Delete Object / Bucket Goal: Use the aliyun command uniformly, minimizing dependency on ossutil.

Rule 4: If Aliyun CLI is installed, ossutil is not needed

This skill does not require ossutil to be installed by default. As long as Aliyun CLI >= 3.3.1 is installed and the following has been executed:

aliyun configure set --auto-plugin-install true

It can be used as the default execution tool.

Core Workflows

Task 1: Create Bucket and Upload Files

Always confirm with the user before creating a bucket. Proceed only after the user agrees.

# 1.1 Create Bucket
aliyun ossutil api put-bucket --bucket examplebucket --region <region-id> --user-agent AlibabaCloud-Agent-Skills
# 1.2 Download files
aliyun ossutil cp oss://example-bucket/test_medias/ /tmp/test_medias_download/ -r --region cn-hangzhou --user-agent AlibabaCloud-Agent-Skills
# 1.3 Upload files
aliyun ossutil cp /tmp/test_medias_download/ oss://example-bucket/test_medias/ -r --region cn-hangzhou --user-agent AlibabaCloud-Agent-Skills

Task 2: Enable Vector Search & AI Content Awareness (Python script or SDK only)

WARNING: You MUST use python scripts/open_metaquery.py to open MetaQuery. Using aliyun ossutil api open-meta-query is STRICTLY PROHIBITED (it cannot configure WorkflowParameters, which prevents enabling AI Content Awareness features ImageInsightEnable and VideoInsightEnable, severely degrading semantic search quality).

Using the Python Script (Mandatory)

Before executing the Python script, complete the following environment setup:

1. Install Python dependencies:

pip install oss2==2.19.1 alibabacloud-credentials==1.0.8

2. Configure credentials: The Python script uses the alibabacloud-credentials default credential chain to automatically discover credentials (supporting environment variables, ~/.aliyun/config.json, ECS instance roles, etc.). No explicit AK/SK handling is needed in the code. Ensure credentials are configured via the aliyun configure command.

3. Verify RAM permissions: Users must have the minimum RAM permissions required for MetaQuery. See references/ram-policies.md. If the user encounters an AccessDenied error, check that RAM permissions are correctly configured.

Enablement Process:

Prepare the Bucket: a. If the user requests creating a new Bucket:
- Run aliyun ossutil api put-bucket --bucket examplebucket --user-agent AlibabaCloud-Agent-Skills with the user-specified bucket name
- If creation fails with a BucketAlreadyExists error:
  - Immediately stop the operation
  - Inform the user: "The Bucket name <bucket-name> is already taken (it may have been created by you or another user)"
  - You MUST ask the user: "Would you like to: 1) Use this existing bucket? or 2) Choose a new bucket name?"
  - Wait for the user's explicit response before continuing. Do not modify the bucket name or use the existing bucket without permission. b. If the user provides an existing bucket:
- First verify the bucket exists using aliyun ossutil api get-bucket-info --bucket <bucket-name> --user-agent AlibabaCloud-Agent-Skills
- If it does not exist, ask the user whether to create it
Verify Bucket object count: After the user provides a bucket, check the object count. If it exceeds 1000, warn the user that enabling MetaQuery will incur costs. Use the following command to get the bucket's object count:
```
aliyun ossutil api get-bucket-stat --bucket <your-bucket-name> --output-format json --user-agent AlibabaCloud-Agent-Skills
```
The ObjectCount field in the response indicates the number of objects.
- If the object count exceeds 1000, warn the user that enabling MetaQuery will incur costs and confirm whether to proceed.
- If the object count is 0, ask the user which files to upload. Upload command:
```
aliyun ossutil api put-object --bucket <your-bucket-name> --key <object-key> --body file://<local-file-path> --user-agent AlibabaCloud-Agent-Skills
```
Run the Python script: After the above steps are complete, attempt to open MetaQuery using the Python script. Python script example:

python scripts/open_metaquery.py --region <your-region> --bucket <your-bucket-name> --endpoint <your-endpoint>

Troubleshooting MetaQuery Enablement Issues

Use the get-meta-query-status command to check MetaQuery status:

aliyun ossutil api get-meta-query-status --bucket <your-bucket-name> --output-format json --user-agent AlibabaCloud-Agent-Skills

Based on the returned status:

Status is Deleted: MetaQuery is being closed. The user should retry later.
Status is Running or Ready: MetaQuery has already been created. Check the following two conditions:
- Whether MetaQueryMode is semantic
- Whether WorkflowParameters contains the following configuration:
```
<WorkflowParameters>
  <WorkflowParameter><Name>ImageInsightEnable</Name><Value>True</Value></WorkflowParameter>
  <WorkflowParameter><Name>VideoInsightEnable</Name><Value>True</Value></WorkflowParameter>
</WorkflowParameters>
```
If MetaQueryMode=semantic and both VideoInsightEnable and ImageInsightEnable are True, the user has successfully enabled MetaQuery in vector mode with content awareness (which greatly improves semantic search quality). No further action is needed. If these conditions are not met, recommend the user switch to a different bucket and start over.

Task 3: Execute Semantic Search

Prerequisites for MetaQuery Search

Before using MetaQuery for search, confirm the following:

Verify MetaQuery is enabled:

aliyun ossutil api get-meta-query-status --bucket <your-bucket-name> --output-format json --user-agent AlibabaCloud-Agent-Skills

If MetaQuery is not enabled: Complete the enablement process first. Refer to Task 2 to enable it using the Python script.
Check index scan status: The Phase field from get-meta-query-status indicates the current scan phase:
- FullScanning: Full scan in progress. Search is not available yet. Wait for the full scan to complete.
- IncrementalScanning: Incremental scan in progress. The index has been largely built and search can be performed normally.
Verify MetaQuery state is Running: MetaQuery is only available when State is Running. If the state is Ready or any non-Running state, you may need to wait or re-enable it.

1. Prepare the meta-query.xml file: Create a meta-query.xml file to define query conditions. For detailed format, field descriptions, and complete examples, see references/metaquery.md. Example of semantic vector search for video files containing "person" (MediaTypes can only be one of: video, image, audio, document):

<MetaQuery>
<MediaTypes><MediaType>video</MediaType></MediaTypes>
<Query>person</Query>
</MetaQuery>

Example of scalar search where file size > 30B and file modification time > 2025-06-03T09:20:47.999Z:

<MetaQuery>
<Query>{"SubQueries":[{"Field":"Size","Value":"30","Operation":"gt"},{"Field":"FileModifiedTime","Value":"2025-06-03T09:20:47.999Z","Operation":"gt"}],"Operation":"and"}</Query>
</MetaQuery>

2. Execute the search command: This example uses semantic vector search. The meta-query.xml file defines the query conditions, and search results return the most similar files.

aliyun ossutil api do-meta-query --bucket <bucket-name> --meta-query file://meta-query.xml --meta-query-mode semantic --user-agent AlibabaCloud-Agent-Skills

For scalar search, use --meta-query-mode basic

For detailed command parameters, see the DoMetaQuery section in references/related-apis.md.

3. Optimizing search result display: After search completes, when displaying results to the user, use the x-oss-process parameter to generate preview images or cover frames for image and video files, making it easier for the user to visually review search results. If the user's current channel supports multimedia files, send them directly to the user.

Video files -- Get video cover snapshot:

aliyun ossutil presign oss://<bucket-name>/<video-object-key> --query-param x-oss-process=video/snapshot,t_0,f_png,w_0,h_0 --user-agent AlibabaCloud-Agent-Skills

Parameters: t_0: Capture frame at 0ms as cover; f_png: Output format PNG; w_0,h_0: Width/height 0 means original resolution.

Image files -- Get image preview link:

aliyun ossutil presign oss://<bucket-name>/<image-object-key> --user-agent AlibabaCloud-Agent-Skills

Note: The aliyun ossutil presign command generates a signed temporary access URL that can be opened directly in a browser for preview during its validity period. For image files, you can also add image processing parameters via x-oss-process (e.g., resize, crop):
aliyun ossutil presign oss://<bucket-name>/<image-object-key> --query-param x-oss-process=image/resize,w_200 --user-agent AlibabaCloud-Agent-Skills
This generates a thumbnail preview to reduce loading time.

Troubleshooting MetaQuery Search Issues

User asks "Why wasn't a specific file found?"

When a user reports that a specific uploaded file is missing from search results, troubleshoot based on the MetaQuery configuration:

a. Content awareness is NOT enabled: If the user's MetaQuery does not have content awareness enabled (i.e., VideoInsightEnable or ImageInsightEnable is not True in WorkflowParameters), possible reasons include:

The file's metadata index has not been fully built yet. Wait for the index scan to complete (check the Phase field via aliyun ossutil api get-meta-query-status --bucket <bucket-name> --output-format json --user-agent AlibabaCloud-Agent-Skills).
Without content awareness, search is based only on basic file metadata (filename, size, type, etc.) and cannot perform semantic understanding of file contents, resulting in limited search effectiveness.
Recommendation: Suggest the user enable content awareness to improve search quality. Since existing MetaQuery configurations cannot be directly modified, recommend the user switch to a new bucket and re-enable MetaQuery with content awareness following the Task 2 process.

b. Content awareness IS enabled: If the user's MetaQuery has content awareness enabled but a specific file still cannot be found, possible reasons include:

File is still being processed: Content awareness requires deep analysis of files (e.g., image recognition, video understanding), which takes longer, especially for video files. Check the Phase field via aliyun ossutil api get-meta-query-status --bucket <bucket-name> --output-format json --user-agent AlibabaCloud-Agent-Skills:
- FullScanning: The overall index is still in full scan mode. Wait patiently.
- IncrementalScanning: Newly uploaded files are being processed incrementally. Usually wait a few minutes.
Unsupported file format: Some file formats may not be supported by content awareness. In this case, search can only use basic metadata.
Search keywords don't match: The user's search keywords may not semantically match the file content. Suggest the user try adjusting their search keywords to use descriptions closer to the actual file content.

Task 4: Query Data Index Status (aliyun ossutil)

aliyun ossutil api get-meta-query-status --bucket <bucket-name> --output-format json --user-agent AlibabaCloud-Agent-Skills

For detailed descriptions of returned fields (State, Phase, MetaQueryMode, etc.), see the GetMetaQueryStatus section in references/related-apis.md.

Verification

See references/verification-method.md

Resource Cleanup

# Close the data index. (Dangerous operation -- confirm with the user first)
aliyun ossutil api close-meta-query --bucket <bucket-name> --user-agent AlibabaCloud-Agent-Skills

Warning: After closing the data index, all indexed data will be cleared. (Dangerous operation -- confirm with the user first)

Alternative Python Scripts for OSS Operations

When aliyun ossutil is unavailable, you can use Python scripts as alternatives. See the Python SDK Scripts section in references/related-apis.md.

alibabacloud-oss-manage-metaquery