Sidhartha Mani

Co-founder, | prev. Apple Containers, MinIO, Kubernetes, Docker

[Sidhartha Mani @utter_babbage]

Getting your first sandbox is easy. Getting thousands of them isn't.

On Thursday I'm showing what we built to fix that: one API for every provider, with routing and failover built in.

Thursday, 10AM PT
https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

Every sandbox provider has a bad night eventually. If your agents hard-code one provider, its outage becomes your outage.

That's why we built https://t.co/eVvkwN9OPZ to route each workload to the right sandbox.

Launching tomorrow at 10AM PT: https://t.co/ughudggtbv

[Sidhartha Mani @utter_babbage]

@clampdown19 Useful split. For the running-code-safely bucket, https://t.co/eVvkwNamFx helps teams find the right sandbox for any workload.

We launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@GoogleAIStudio Love this!

We'll integrate so you can bring your own sandbox via https://t.co/eVvkwNamFx, and run Gemini agents on whichever provider fits the workload.

If you're reading this, join us for the launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@arjunraime Thanks Arjun, appreciate the hype! 🙌

[Sidhartha Mani @utter_babbage]

@MenloVentures Agreed, getting one sandbox is easy, but getting thousands in a burst is a different problem.

We're launching https://t.co/eVvkwNamFx to help teams find the right sandbox for any workload.

Would love to have your support at our launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@SvenMeyer @harshrajdhanke @morganlinton @nvidia @vercel Cost swings a lot from one workload to the next. We can help you find sandboxes at a budget.

Check out https://t.co/eVvkwNamFx and join us for the launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@pavitrabhalla @amyfriar_ Ha, thanks Pavitra! Always fun seeing the numbers come in. Appreciate you being part of it 🙌

[Ivan Burazin @ivanburazin]

Only ways to solve the CPU crunch simultaneously:
1/ build more
2/ optimize utilization of what already exists to fit more workloads into the same hardware

Right now none of them is moving fast enough to meet the demand that will follow once consumer apps hit scale.

[Sidhartha Mani @utter_babbage]

@anuraggoel @render Love this. Agents picking the right sandbox per job is where this is headed.

Would love to have Render in the mix for https://t.co/eVvkwNamFx. Join us for the launch tomorrow: https://t.co/ughudgh113

[Maria Gorskikh @mariagorskikh]

we ran a simulation with 32 agents on 32 computers playing chess against each other

16 multiplayer games at once, every move clicked on one computer and shown on the opponent's

2,000 moves later we got our champion!

[Sidhartha Mani @utter_babbage]

@mernit Very cool. Would love to have Beam in the mix for https://t.co/eVvkwNamFx, GPU sandboxes are a big part of where agent workloads are going.

Join us for the launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@michielmv @ag2oss Congrats on the AG2 integration! Would love to have boxd in the mix for https://t.co/eVvkwNamFx.

Join us for the launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@alxshp Agreed, with state outside the sandbox, the sandbox can run anywhere.

Check out https://t.co/eVvkwNamFx and join us for the launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@0xblacklight Agreed the harness will own orchestration. The sandbox underneath should stay swappable though, so each workload can run where it fits best.

That's what we're building at https://t.co/eVvkwNamFx. Join us for the launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@IgorIlyinsky @michael_chomsky Agreed, that's the layer we've been building. Each run should land on whichever sandbox fits it best. Join us for the launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

@digitalocean @paddix @theinformation Agreed, and the right sandbox depends on the workload. We built https://t.co/eVvkwNamFx to help teams find the right sandbox for any workload.

[BoxLite @BoxLiteAI]

BoxLite is a masterpiece.

Most AI sandbox discussions focus on speed and pricing.

But there’s another question:

Where should your agents actually run?

With BoxLite, developers can embed hardware-isolated microVMs directly into their applications.

> Open source.
> Local-first runtime.
> Built for untrusted code execution.

Not every agent needs another cloud dependency.

[Sidhartha Mani @utter_babbage]

@an_engineer_log Would love to have Agent37 in the mix for https://t.co/eVvkwNamFx. More options for teams to pick the right sandbox for each workload.

Join us for the launch tomorrow: https://t.co/ughudgh113

[Sidhartha Mani @utter_babbage]

Agreed, the VM is the easy part and state is the hard part. Block-level snapshots can't tell when app state is whole, so we think the industry moves to app-level snapshots.

That's part of what we're building at https://t.co/W037pPwqSJ. We launch tomorrow, would love to have you there: https://t.co/BecToQxWLR