Urbit

This is v0.1. Open, hackable, and already doing things Alexa can't — local LLM inference, real-time sound classification, full developer access. Every product starts somewhere. This is ours.

It's so liberating to go back to simple straightforward push deployments. Like old-school Ansible over SSH but feeling more like a modern container orchestrator. It took me a decade to come full circle. If you've been around long enough, you probably remember the evolution: configuration management tools, then the Docker boom, followed by the race of container orchestrators. K8s eventually won and we all agreed that the correct way to deploy software is: 1. Declare desired state (I want to run this service, don’t care how) 2. Let background controllers make it a reality continuously 3. Hope nothing silently fails 4. Set up a dozen different alerts so we actually know when it does 5. Every failure means tracing numerous resources, events, and logs to figure out the actual chain of actions and the root cause 6. And now adding more fuel into the fire - AI agents to help keep up with all this That’s the current norm. One solution that fits all. Sadly many engineers don't even know alternatives or question this because that’s what they all do. Reconciliation loops are brilliant engineering. At massive scale, you do need these autonomous controllers pulling and converging state and healing the dynamic system. It’s impractical to build a reliable synchronous workflow that would reliably manage services on thousands of machines over SSH. But the main tradeoff is poor visibility and troubleshooting. We can do better for smaller scale - which is most of the web apps tbf. But we need to go back and revisit the choices, not put more duct tape. We can keep the declarative definitions and reconciliation but run it synchronously on the client, not in a background loop you have no visibility into. You run the deploy, you see the result, you get a clear error if it fails. You instantly see what went wrong and what to troubleshoot or retry if needed. Your automated pipeline could do this too of course. And so this is how we designed Uncloud - push-based deployments with the lessons from the orchestrator era. Check out 👇

Her bank knew she was pregnant before she did. A statistician at Target built a model in the early 2000s to identify pregnant customers through purchase patterns. The algorithm predicted due dates within a few weeks. Target mailed pregnancy coupons to customers who'd never disclosed a pregnancy. One father in Minnesota got the mailer, confronted his teenage daughter and called Target. She was pregnant. He didn't know. The algorithm did. This was 2002. Before smartphones. Before behavioral biometrics. Before machine learning was a commodity. Before your bank had 20 years of granular transaction data, before insurance algorithms, before fintech data brokers who buy and resell your financial patterns to anyone with a budget. Think about what is possible today. Your stress levels - pharmacy purchases spike before major life events. Your relationship status - spending shifts at predictable milestones. Your political leanings - donations, subscriptions, events attended. Your health conditions - specialty pharmacies, supplement patterns, clinic co-pays. Your financial fragility - overdraft timing, the exact rhythm of living paycheck to paycheck. None of this is hypothetical. All of it is for sale. The Target story got press because it was awkward. What doesn't get press is the infrastructure that makes it routine. Let there be Dark!

Privacy tier list