Video yükleniyor...
Video Yüklenemedi
Introducing Rivet 3.0 (preview) Built on S3 with SSD speeds 🏎️ 8ms actor coldstarts 🪶 0.7ms SQLite writes 🤏 78 KB of RAM per actor 🍱 Embed, self-host, or BYOC 📁 S3 is the only dependency, no database
51,396 görüntüleme • 22 gün önce •via X (Twitter)
33 Yorum

We built WarpDrive for Rivet 3.0, our new storage medium backing Actors & SQLite. Every actor gets bottomless storage on S3 with the speed of an SSD, and keeps running even if S3 goes down.

🧰 Every actor has memory, storage, and compute ⚡️ Sleeps when idle and scales to billions 🏢 Run thousands more agents on the same hardware

Already on Rivet? No code changes needed. On Rivet Cloud, you only pay for hardware: ❌ No awake actor hours ❌ No read/write fees

Read more:

GitHub:

Discord:

fire stuff 🔥🔥. at this point s3 will be the last software.

@jidefr @NathanFlurry very clear message, very cool product. Bravo

can’t wait to self host this bad boy on my unlimited Verizon cloud plan

At the 0.7ms write acknowledgement, where is the data durable? Local disk, another node, or S3? That's the first detail I'd want before comparing it with a database write.

thats my goat @NathanFlurry @NicholasKissel

🥳

nice one. congrats but wondering how parking it on mem works in a way I’m not paying for idle times

i need to move more of my stuff to rivet.

@bot this seems really powerful!

8ms to wake up feels rude to the rest of us

congrats guys!!

S3 with SSD speeds, that's the claim to watch.

This is really cool I believe so much in this way of doing agent envs

are the coldstart numbers from a clean load off s3 or does it keep a warm pool behind it

S3 as the only dependency is a bold flex. If it really feels like SSD speeds, a lot of infra complexity just became optional.

can somebody explain me what problem it is solving?

insane

@grok 这个有什么用?

Not Open sourced with our own clouds?

Love it! This is the perfect architecture for deploying actors really love the change. I read the article, how do we migrate now? Do we have a guide if we don't want wait for the automigrate tool?

A nice follow-up demo would be an actor waking after a week idle and querying a large SQLite dataset. Cold-start time plus first useful query time would make the storage story much easier to picture.

This is INSANE

lfggg

will @AmpCode implement? :D

are those 0.7ms sqlite commits flushing to s3 synchronously or batched in memory

That’s interesting. 🔥

This is insane 🔥🔥🔥



