Rebecca

Managing Partner @ MAXIMIZE - developer marketing, formerly Sponsored, @MongoDB, @DigitalOcean

[Rebecca @beckfastattiffs]

@glcst But isn’t it fun saying “I told you so” when the rest of the world catches up?

[Glauber Costa @glcst]

I am excited to announce that we are officially writing a new version of Postgres. In Rust - and creating the LLVM of databases in the process.

In the span of a year, we have rewritten SQLite. Keeping the compatibility, increasing its feature set. MVCC, Types, (Live) Materialized Views, among other things. In the process of doing that, we have realized: At the end of the day, what makes SQLite special is that it compiles SQL to a database-specific bytecode. So why can't we compile *Postgres* to the same bytecode?

Turns out we can. I ran an experiment called pgmicro as a way to prove this hypothesis, and it works very well. It is time to make this official, and put the weight of Turso behind it. We shall give the world a modern take on Postgres. Wire compatible, but built on a new architecture.

We have already heard of others wanting to extend this. MySQL? Redis? the sky is the limit. What can we do if we do for databases what LLVM did for compilers? To prove how powerful the SQLite bytecode is, we are actually running DOOM compiled to the unmodified SQLite instruction set. And because Turso runs natively in the browser, you can play the game in your browser. With the database executing it.

Read the full story below! 👇

[Glauber Costa @glcst]

Weekly Active Users on the Turso Cloud

If you are one of the users now finding what a lighweight database can do, either locally or in the Cloud... welcome!

[JS0N Haddix @Jhaddix]

Excited to join Rubrik next week for a conversation on resilience, AI, and what modern enterprises need to be thinking about next.

https://t.co/Aqr2PDWO59

[Blaxel @blaxelAI]

Blaxel is now integrated with Stripe Projects!

Now, agents (and humans) can spin up accounts and resources from the terminal directly through @Stripe, and get back credentials without dashboard hopping.

https://t.co/hTiOfb6con

[Glauber Costa @glcst]

Ok, trying this again, since last week I had trouble convincing the Reddit mods that I am me.

Tomorrow, 9 AM Central: ASK ME ANYTHING! (you can start asking now, and I will reply tomorrow)

https://t.co/2ajYKjYSBY

[Mitch Wainer @MitchWainer]

So proud of our client, Hexo Labs, as they announce their open source self improving AI called SIA. Superintelligence is becoming a closer reality to solve the worlds most challenging problems.

[Alvaro Cintas @dr_cintas]

HexoAI just open-sourced a self-improving AI that updates its own weights 🤯

The loop runs three agents:

> a meta-agent builds an agent for your task
> the target agent attempts it and logs everything
> a feedback agent rewrites the harness and updates the weights

Then it repeats, improving each generation.

100% Open Source.

[Rebecca @beckfastattiffs]

@kunalbhatia91 @hexoai Incredible achievement! Congrats to the entire team!

[Kunal Bhatia @kunalbhatia91]

Weights update is the real breakthrough in continual learning

Frontier coding agents are strong, but they're frozen. Point Claude Code or Codex at a task and they can't keep getting better at it. You can see it across all three benchmarks: they sit near the baseline and barely reach the prior SOTA line.

With harness-only updates, we land in roughly the same neighbourhood.

The breakaway happens once you update the model's weights, far past everything else. Self-improvement isn't a smarter scaffold. It's letting the model actually learn.

In a harness update, the meta-agent only teaches the task-specific agent software engineering i.e. better parsers, retry logic, tool dispatch, search procedure. It never touches the domain itself.

On LawBench it can build a cleaner classification pipeline, but it can't make the model understand Chinese criminal law.

Weight updates do exactly that: gradient pressure pushes the model into latent reasoning about the problem - disambiguating 191 charge categories, internalizing H100 kernel patterns, learning that imputed RNA counts must be non-negative integers.

The harness shapes how the agent searches; the weights make it a domain expert.

[Kunal Bhatia @kunalbhatia91]

Superintelligence will be built on Self Improvement.

Today @hexoai, we’re excited to release ‘SIA’ - an open-source Self-Improving AI, to achieve any goal through recursive self improvement.

While trying to solve a problem, SIA doesn't just improve it's abilities by updating it's harness, it updates it's own weights as well.

[Glauber Costa @glcst]

Want to know more about how Turso came to be? I am doing an AMA on Reddit tomorrow - 8 am Pacific / 10 am Central, 11 am Eastern

Stay tuned !

[Kate Tolo @_katetolo]

i too lost my boyfriend to claude

[Glauber Costa @glcst]

Today we are announcing Turso for startups, our startup program that helps you take advantage of the many-database architecture: https://t.co/RsZ9kyceJE

The rise of agents made this obvious: lightweight, isolated databases that come online in milliseconds and can be used anywhere are the architecture of the next 10 years.

Whether you want a database per user, per agent, or even per session, @tursodatabase is the database for you: the only provider that allows you to spin up millions of databases and pay only for what you use.

If you are a VC-backed startup backed by a reputable fund, we are now offering you credits to build safe and scalable AI architectures on Turso

[sui ☄️ @birdabo]

everybody calm down.

i got this.

[Alexander Whedon @alex_whedon]

Introducing SubQ - a major breakthrough in LLM intelligence.

It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),

And the first frontier model with a 12 million token context window which is:

- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus

Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).

Only a small fraction actually matter.

@subquadratic finds and focuses only on the ones that do.

That's nearly 1,000x less compute and a new way for LLMs to scale.

[Glauber Costa @glcst]

Turso now includes unlimited active databases in every plan. We already had unlimited databases, but we would charge you based on how many of them were active. That is now gone. You want a database, you get a database.

[Glauber Costa @glcst]

We are now letting the first couple of people in for our private beta. If you haven't received anything from us yet, stay tuned!

And if you want to deploy millions of databases to power your agents, with concurrent writes and all the goodies that Turso offer - securely, privately, with the option of deploying on our own cloud account, join the waitilist now!

https://t.co/c5ejeLFp60

[Blaxel @blaxelAI]

Blaxel is officially a first-class sandbox provider in the new version of @OpenAI Agents SDK.

Your agents now have shell, file I/O, and live preview URLs, all running in isolated cloud sandboxes on Blaxel infra.

https://t.co/7sPjOmNEWk

[Glauber Costa @glcst]

We at @tursodatabase are ecstatic to announce today the private beta of the new generation of the Turso Cloud.

The Turso Cloud has been a great destination for builders building agents, that recognize that agents need millions of databases, and that the SQLite file-based approach is the way to achieve that at scale.

But at the same time, SQLite comes packed with limitations. Over a year ago, @penberg had a bright and bold idea - perhaps the only good idea he had in his life - what if we rewrote SQLite entirely? If we do that, our Cloud can keep the unmatched database density at scale it has today, but without that ceiling that makes SQLite be seen as a temporary solution.

Fast forward to today, Turso, our rewrite of SQLite, is right around the corner. Concurrent Writes are here, and many other features.

And select teams will be able to run it first in the Turso Cloud: query it directly, or use it to sync your Turso databases. All of that, with enterprise-grade encryption, and running in your own cloud account.

Link 👇