OSS Agent is an agent service provided by Alibaba Cloud Object Storage Service (OSS). It lets you use natural language to manage buckets, diagnose anomalies, perform health checks, analyze costs, and semantically retrieve and interact with multimodal data, such as videos, images, and documents, stored in your buckets.
OSS Agent is currently in an invitation-only preview. To request access, contact Technical Support.
Overview
OSS Agent is built on the Tongyi large model. It understands natural language intent, automatically plans and executes multi-step API call chains, and visualizes the process and results in the workspace. OSS Agent provides the following two core capabilities:
intelligent O&M: Use natural language to perform O&M tasks such as bucket configuration, access anomaly diagnosis, health checks, and cost analysis. This capability is ideal for scenarios that require rapid multi-step configurations, troubleshooting access issues, or periodically assessing bucket health.
Talk to Bucket: Powered by the content-aware capabilities of OSS Metaquery, this feature lets you retrieve and interact with multimodal data (such as videos, images, audio, and documents) in your buckets using natural language, without needing to remember file paths or construct query statements.
All operations are managed and categorized by risk level: low-risk read operations are executed directly; medium-risk write operations require your confirmation before execution. The agent does not directly execute high-risk operations, such as deletions and lifecycle management. Instead, it provides console or API/SDK procedures to follow.
Prerequisites
You have activated OSS.
If you use a RAM user to access OSS Agent, ensure that the user has the required permissions for the target bucket.
Use OSS Agent
Log on to the OSS console and go to the OSS Agent conversation page.
On your first visit, read and agree to the Terms of Use to access the chat interface.
In the input box at the bottom of the home page, enter your request or select a question from the Intelligent O&M Expert, Q&A Assistant, or Chat with Multimodal Data cards to get started.
The OSS Agent interface consists of the following three areas:
Left sidebar: Provides access to new sessions, Region and bucket selectors, the session history list, and the skill management entry.
Main chat area: Interact with OSS Agent by entering your requests. You can use
@in the input box to temporarily reference a bucket and specify it for an operation.Workspace: Acts as a collaborative area between the agent and the console, and provides built-in tools such as Create Bucket, Resource Plan Recommendation, and Error Code Diagnosis.
Intelligent O&M
Multi-step operation orchestration
For operations that require multiple configuration steps, OSS Agent automatically breaks down natural language commands into an ordered API call chain, with the result of each step displayed in real time. For steps that involve costs or risks, the agent prompts you for confirmation before execution.
Example: Create a bucket named sample-bucket in the China (Shanghai) region, enable server-side encryption and versioning, and configure a permission policy that allows only internal network access.
Agent execution chain: ① Create a bucket in the Shanghai region → ② Enable server-side encryption (AES256) → ③ Enable versioning
(Enabled) → ④ Configure a bucket policy to restrict access to only the internal network (VPC).
For parameter-sensitive operations, such as creating a bucket, the agent passes the recommended parameters to the native console form, which is displayed in the workspace for your review and modification. The operation is executed only after you confirm.
Access anomaly diagnosis
Provide the URL of an inaccessible object or an error message to OSS Agent. The agent automatically initiates a multi-step diagnosis, identifies the root cause, and provides recommended fixes.
Example: I can't access this image. Can you tell me why? https://examplebucket.oss-cn-hangzhou.aliyuncs.com/exampleobject?x-oss-additional-headers=host&x-oss-credential=LTAI********************%2F20241203%2Fcn-hangzhou%2Foss%2Faliyun_v4_request&x-oss-date=20241203T034420Z&x-oss-expires=3600&x-oss-signature=70c542eaf652ac291c0c343d63ac24ede41c0526661d9d4c63c0906a**********&x-oss-signature-version=OSS4-HMAC-SHA256
The agent identifies the root cause (for example, "This is an OSS V4 presigned URL that was valid for 3,600 seconds, and the signature has now expired") and provides steps to resolve the issue.
Bucket health check
Initiate a health check on a specific bucket. The agent automatically inspects four dimensions: security, data protection, cost management, and basic information. It outputs the status and a description for each item and suggests optimizations for any identified risks. Upon confirmation, the agent can perform the corrective actions.
Example: Perform a health check on my bucket @samplebucketname.
The agent automatically executes multiple API calls to cover the four health check dimensions.
Dimension | Check item |
Security check | Bucket ACL, server-side encryption, block public access, hotlink protection, CORS |
Data protection | versioning, delete markers, multipart upload part cleanup |
Cost management | lifecycle rules, access logging, access monitoring |
Basic information | bucket overview (region, storage class, redundancy policy, object count, storage usage) |
Cost analysis
The agent automatically retrieves billing data and analyzes it by billable item, such as storage fees, data transfer fees, and request fees. It then outputs a structured cost report to help you understand the cost breakdown for each bucket.
Example: Help me analyze my costs.
Talk to Bucket
How it works
The Talk to Bucket capability of OSS Agent is powered by OSS Metaquery data indexing, which uses AI content awareness and vector search. It uses a three-layer architecture to process semantic searches:
Content awareness layer: Uses multimodal large models to understand the content of videos, images, audio, and documents, generating semantic descriptions and structured tags.
Vector indexing layer: Converts the content analysis results into high-dimensional vectors to build a vector index.
Semantic matching layer: Transforms the user's natural language input into a query embedding, performs a multi-path search and re-ranking in the vector database, and returns the top-K matching results.
The entire process is transparent to the user. OSS Agent guides you through the full workflow, from enabling MetaQuery data indexing and vectorizing existing data to performing searches and conversing in natural language.
Enable data indexing
In the floating panel at the lower-left corner of the input box, toggle the Talk to Bucket switch to start a session. When enabled, your messages automatically trigger the multimodal data retrieval skill to perform intelligent searches and answer questions about the selected bucket. When you use the Talk to Bucket feature, OSS Agent automatically checks whether MetaQuery data indexing is enabled for the target bucket and guides you to enable it if necessary.
Enabling MetaQuery data indexing incurs fees for AI content awareness and semantic vector search. Before enabling the feature, OSS Agent displays a cost notification. Confirm the costs before you proceed. For more information about MetaQuery billing, see Data Indexing Fees.
Vectorizing existing data takes some time, and searches can be performed only after this process is complete. During this period, OSS Agent displays the current indexing status.
Use cases
Semantic image search
This capability is ideal for scenarios such as media asset libraries. You can search for images by describing their content, without relying on manual tags or file naming conventions.
Example: @your-bucket-name Find me pictures of a volcanic eruption.
The agent returns matching images, each with its file path, size, resolution, tags, and a description.
Direct interaction with multimodal data
This capability is suitable for scenarios such as security surveillance and video archives. You can locate specific clips within massive video collections by describing an event, eliminating the need to review footage segment by segment.
Example: Can you find out who caused the fire? If you find a video, please send it to me.
The agent returns matching videos, along with the video path, duration, an AI-generated scene description, object detection results, and a presigned URL for direct download. The URL is valid for 15 minutes and must be regenerated if it expires.
The results from AI content awareness are for reference only. Accuracy is affected by factors such as video clarity and cannot replace manual review.
Semantic document search
This capability is suitable for scenarios such as enterprise knowledge bases and technical document management. The agent reads document content and returns relevant documents and access links based on semantic matching, rather than relying on file names or paths.
Example: Find me documents related to a certain internet platform project.
Quotas and limits
Item | Description |
Concurrent sessions | A single UID is limited to two active sessions on one device at a time. |
Session lifecycle | A session is automatically terminated if it is idle for more than 1 hour or if the cumulative conversation time exceeds 8 hours. |
High-risk operation limit | The agent does not directly execute high-risk operations. These include deletion operations (for buckets, objects, and configurations), lifecycle management, write-once-read-many (WORM), data replication, overwrites, and static website hosting. For these operations, the agent provides a link to the OSS console. |
Billing
OSS Agent is currently in a free, invitation-only preview. However, operations triggered through OSS Agent that use other OSS features, such as enabling MetaQuery data indexing or accessing data, are billed according to the corresponding OSS pricing. The agent notifies you of potential costs before executing such operations. For more information, see Billing overview.