AWS S3 · Connector
How do we connect an S3 bucket to AI agents on our own infrastructure?
Katonic connects to an S3 bucket, reads the files under the prefix you name and indexes them in the Knowledge Engine on infrastructure you control. Agents answer from those files with a model you serve, so the files and the questions are not sent to a public AI service.
What it reads.
From AWS S3.
- 01The files in the bucket you name
- 02Only the files under a prefix, where you give one
- 03Only the file types you choose, where you list them
Through: The AWS S3 API
Where the content goes
Into a knowledge store you run.Content is pulled into the Knowledge Engine on infrastructure you own or control, and indexed there. Agents answer from it with a model you serve. The documents, the index and the questions are not sent to a public AI service.
How it signs in
An access key ID and a secret access key. The connector also supports an IAM role.
The set-up dialog asks for
- Bucket name
- AWS region
- Access key ID
- Secret access key
- Prefix or path (optional)
Permissions
Access is set on the source in Katonic: everyone given access to the source can retrieve its documents. The permissions AWS S3 holds for each item are not read.
An S3 object has no permissions that map to your users and groups.
Sync
A full listing of the bucket at each sync, reading the files changed since the last one.
You set how often the source is read again, in minutes, hours, days or weeks.
Limits
- The connector itself works with S3-compatible stores such as MinIO, through a custom endpoint setting. The set-up dialog asks for an AWS region and has no field for a custom endpoint today.
- A store on a private network address needs an operator to add it to an allowlist first.
Does it work with MinIO or another S3-compatible store?
The connector supports a custom endpoint for S3-compatible stores. The set-up dialog does not ask for one today.
Can we limit it to one folder?
Yes. You give a prefix, and only the files under it are read.
Where are the files indexed?
In the Knowledge Engine, on infrastructure you own or control. They are not sent to a public AI service.