Announcing Amazon S3 Files. The first and only cloud object store with fully-featured, high-performance file system access. Learn more here.

Think about what this means for agentic AI. Every coding agent, every data pipeline agent, every automation tool that shells out to the filesystem, they all just got native S3 access without anyone writing a single line of integration code. No SDK, no API wrapper, no staging layer. Just mount and go. This quietly removes one of the biggest friction points in building AI systems that work with real data at scale.

So now, instead of sandboxes, you can point a swarm of agents directly to the filesystem that already has ALL your data. This is massive… I am extremely excited to play with this.

Here’s my take from @VentureBeat on this :

Agents LOVE directory structures. I think this is indirectly a huge win for agent harnesses

native file access for s3 is a massive win. finally.

Can’t wait to point my LLM wiki to be backed by S3

so like ?

so now you can mount s3 bucket to a gpu?

@grok is something changed to s3 ?

@CloudflareDev 👀

This is a massive update! 🚀 High-performance file system access directly on S3? Amazon S3 Files is going to simplify so many cloud architectures. Game-changer! ☁️📦 #AWS #AmazonS3 #CloudComputing #Storage

Use case example?

Thank you

It is still unclear from the blog post / marketing page / docs is it really "accessible from any Linux-based" or from EC2, ECS, EKS, Fargate, and Lambda only. Say, can I mount it to my MacOS?

@larsencc

This is huge! Finally mounting S3 buckets directly as a proper high-performance filesystem without all the ETL headaches 😍 No more copying data around or dealing with awkward SDKs for agents. Game changer for AI/ML workflows. Well played AWS!

This is awesome, amazing product! Can't wait to use it with Daft to leverage our world record setting S3 reads/writes with: - daft.File lazy file references - daft.functions.prompt - daft.functions.embed

Should my ec2 now just use s3 instead of ebs?

we can use this for deepagent file system access right? instead of normal s3 @Vtrivedy10

Great! So in this way, ECS tasks could work directly with file from S3 bucket, without the need to download it first? Is it going to be released in all regions?

File systems are an abstraction that helped humans organize information in a way other humans understood “natively”. This seems pointless now.

Awesome work 👏

Thank you! 🤩

@archildata are you ok?

this is one of those "wait why didn't this already exist" features. everyone's been hacking file access on top of S3 forever

How is it different to Azure files which I was using 10 years ago.

Impressive! File system access on S3 is a game-changer. We've been building AI agents that work with storage - would love to integrate this. Check out our work at

WOW WOW WOW WOW! LESSGOOOOOOOO HOLY SHIT! Thanks Aws

Someone took hf-mount really personally and close to the heart

I can’t imagine regular users paying to access for their files (egress), this will only work for enterprise

Use S3 Files on @aisdk :)

@Mur_Gain я ничо не пон но тебе точно понравится

@grok how is this different from s3fuse or similar tools?

S3 is just managed CephFS Object Store EBS is just managed CephFS Block Store FSx is just managed CephFS Filesystem You people will buy anything

Incredibly expensive!

any benchmarks with sqlite?