On the side of game development I had some fun and built an entire startup and tech stack: tiderun.ai

It's based on streaming FaaS — serverless architecture with AI as a first-class citizen.

The core question

I wanted to explore how you combine JavaScript runtimes with Rust executables. How would you best serve SSE streaming through FaaS in these interesting times — where Europe is also considering breaking away from US cloud dependency?

RAGs, prompt-cache gateways, all sorts of interesting language model patterns to work with. The tooling landscape is moving fast and the infrastructure needs to keep up.

Streaming as a first-class primitive

Traditional FaaS is request-response. You call a function, it runs, it returns. But AI workloads are fundamentally different — they stream. Token by token, chunk by chunk.

AWS has streaming for their Lambdas but it feels like an afterthought. Streaming isn't a first-class primitive there.

Tiderun is my exploration of what it looks like when you design for streaming from the ground up.

What is it then?

In its simplicity — I built a streaming-first serverless architecture using V8 runtime sandbox isolation at its core. Combined with Rust for the fun and practical benefits of streaming callbacks and memory management.

Think Supabase or Vercel if they were a standalone cloud.

One little function paired to an endpoint, running in isolation but sharing the same runtime container for many functions and calls in parallel. And streaming responses the whole time.

Deno has been doing an amazing job for the Rust community when it comes to managing Google's V8 and supplying an SDK layer to bootstrap and inject all sorts of functionality on top of it. For example, your own internal SDK functions exposed directly to JavaScript but running in Rust. Similar to how AWS and Cloudflare do it under the hood.

The sovereign cloud problem

I have an entire FaaS infrastructure designed for the sovereign cloud. The problem is that I don't own a cloud. And don't want to start one right now.

It's more of an orchestration layer that could be applied to any cloud to have a standalone runtime up and going. And of course an LLM to generate these functions for us. A nice dashboard layer to manage and monitor the infrastructure, inference chat capabilities to create new functions and manage existing ones.

But the question is real and getting more relevant by the day. As Europe considers its relationship with US cloud providers, there's a genuine need for infrastructure that can run independently. Streaming-first serverless is an interesting piece of that puzzle.

...honestly thought

To much politics.. and old IT-men. I rather go back to making games and coding stuff then running a IT Cloud. Imagine all the Kubernetes yaml files I would have to write to deploy this thing.

RIP Docker compose in production.