> For the complete documentation index, see [llms.txt](https://doc.duaer.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.duaer.com/guides/s3-bucket-knowledge-search.md).

# Search files in S3 and Alibaba Cloud OSS buckets with Duaer

Connect the files under a prefix of an S3 bucket, or of Alibaba Cloud OSS, Tencent COS or MinIO, to the Duaer knowledge base. People ask about the policies, manuals and reports in it; everyone in the organization finds them.
## What does Duaer find in a bucket?

Duaer indexes up to 5,000 files under the prefix. PDF, Word, Excel, PowerPoint, text, Markdown and CSV files up to 20 MB are read in full; images and other files are found by name. Files outside the prefix are never listed or read.

People can ask “How are taxis on business trips reimbursed?”. Duaer finds the answer in the expense policy in the bucket and cites it. Files in a bucket have no address that opens without signing, so the citation shows the file name.

## How do you connect a bucket to Duaer?

1. In Alibaba Cloud RAM, Tencent Cloud CAM or AWS IAM, create a read-only access key that may only list and read this bucket (for example oss:ListObjects and oss:GetObject on OSS, s3:ListBucket and s3:GetObject on S3).
2. In Duaer, create a credential: for Alibaba Cloud OSS, Tencent COS or MinIO, an S3 credential with the store’s S3-compatible endpoint (such as https://oss-cn-hangzhou.aliyuncs.com or https://cos.ap-shanghai.myqcloud.com) and its region; for AWS, an AWS (IAM) credential.
3. As an owner or admin, open Data sources in the sidebar, select Add source, choose S3 bucket under Files (a search for OSS finds it too) and Knowledge source, then the credential, the bucket and the prefix (such as handbook/), and click Connect.
4. When indexing is done, ask a question whose answer is in a file in the bucket and check that the citation points to it.

## Who finds the files?

Everyone in the organization. Bucket policies and ACLs do not name Duaer’s people, so Duaer does not tell people apart: connect only a prefix every member may read, and keep sensitive files under another prefix.

## What are the limits?

- One bucket and one prefix per organization, up to 5,000 files; a file over 20 MB is found by name only.
- A direct connection needs an https endpoint; MinIO or another store inside your network is read through an on-premises data gateway that lists it.
- A changed file (its ETag changed) is read again; a deleted one leaves the knowledge base.

To see how Duaer puts knowledge search and follow-through together, open the [Duaer home page](https://www.duaer.com/).
## Questions

### Does Duaer change files in the bucket?

No. Duaer only lists and reads files to index them. It never writes or deletes anything; a read-only access key is enough.

### How soon does Duaer pick up a changed file?

Duaer lists the files under the prefix every 10 minutes and reads again any file whose ETag changed.

### Does Duaer work with Tencent COS and MinIO?

Yes. Both speak S3’s API: choose an S3 credential with their S3-compatible endpoint.

## Related

- [Turn SharePoint and Google Drive into a Q&A knowledge base with Duaer](https://doc.duaer.com/guides/sharepoint-drive-knowledge-base.md)
- [Search Dropbox in your knowledge base with Duaer](https://doc.duaer.com/guides/dropbox-knowledge-search.md)
- [Permission-aware enterprise AI search: how Duaer keeps document access intact](https://doc.duaer.com/guides/permission-aware-ai-search.md)
- [AI answers with citations and no hallucination: how Duaer does it](https://doc.duaer.com/guides/ai-answers-with-citations.md)
- [How to build an AI assistant on your company documents with Duaer (no code)](https://doc.duaer.com/guides/ai-assistant-company-documents.md)