Crusoe For Developers

Gave a talk at our SF infra meetup w/ @dstackai + @sgl_project: 200B+ MoE → resilient multi-GPU inference endpoint on @CrusoeAI Managed K8s. Two commands. KServe + Envoy AI Gateway + vLLM doing the heavy lifting underneath.

Always a great time to learn when @lmsysorg, @dstackai and @CrusoeAI are in the same room.

Hack.Crusoe kicks off today. One week. All Crusoe engineers. Any open model. Build something real on @CrusoeAI Managed Inference and ship it. Demo Day: May 18. #hackcrusoe26

Most agent demos don't survive production. These teams are building the ones that do: → @NousResearch (Hermes) → @emergence_ai (enterprise agents) → @traversal_ai (SRE agents) One room. What actually breaks — and where open models hold up vs. where closed still wins. @CrusoeAI Talks · 📅 May 19 · 📍 Tribeca 🎟️ lnkd.in/gavZwYnP #CrusoeAITalks

@robertshaw21 @dstackai @sgl_project @CrusoeAI Also feel free to checkout our blog in the meantime crusoe.ai/resources/blog…

@CrusoeDev @dstackai @sgl_project @CrusoeAI Slides?

Gave a talk at our SF infra meetup w/ @dstackai + @sgl_project: 200B+ MoE → resilient multi-GPU inference endpoint on @CrusoeAI Managed K8s. Two commands. KServe + Envoy AI Gateway + vLLM doing the heavy lifting underneath.

@YouPulseX @dstackai @sgl_project @CrusoeAI Checkout our blog here: crusoe.ai/resources/blog…

@CrusoeDev @dstackai @sgl_project @CrusoeAI The receipt is the runbook. "Two commands" is the claim people can test. Drop the exact commands, versions, and the boundary between KServe, Envoy AI Gateway, and vLLM.

@DPattnayaak @CrusoeAI This is an internal hackathon. Keep an eye out for our upcoming external ones.

@CrusoeDev @CrusoeAI devpost link ?

$ init hackathon --month may --year 2026 Hack.Crusoe is coming. May 5–11, teams across @CrusoeAI go heads-down building on our own inference platform — one week to ship something real. Stay tuned for what our engineers cook up.

That's DevRel, in their spare time. Next week, the whole company ships.

One example: 🎙️ FakePod — an unscripted AI podcast where Qwen 3 and Nemotron riff on any topic for 20 minutes straight. Real TTS voices, dual-model orchestration, full SSE streaming. huggingface.co/spaces/eacheam…

Small teams can now build multi-model AI systems that would've taken entire orgs a year ago. @CrusoeAI's Hack.Crusoe drops May 5. Ahead of it, our DevRel team got on stage and showed what they've already shipped on Crusoe Managed AI 🧵

@NVIDIAAI @CrusoeAI Excited to dive into Nemotron.

@CrusoeDev @CrusoeAI 🙌 This is great @CrusoeDev - thanks for sharing!

@NVIDIAAI Nemotron 3 Nano Omni is live on @CrusoeAI Intelligence Foundry Open-weight omni model — text + image + video + audio in one sequence. 7.4x multi-doc and 9.2x video throughput vs Qwen3-Omni on B200. Let's build. 🔗 crusoe.ai/resources/blog… Image source: lnkd.in/g6jPZ8hg

Tokenization is quietly eating your TTFT. Our VP of Engineering Omer Landau is breaking it down at Dynamo After Hours on May 6 — an @NVIDIAAI developer event with lightning talks from SGLang, @lmcache, @vllm_project, @Azure, and more. ➡️ luma.com/MayDynamoAfter… #Inference #TTFT

If you're still babysitting training jobs, you're compensating for your infrastructure. When's the last time you got paged for something your system should've handled itself?

A training node dies at epoch 50. [Epoch 50] Loss: 0.0010 *** JOB CANCELLED — NODE FAILURE *** Resuming from epoch 51. [Epoch 51] Loss: 0.0007 No humans in the loop. ~15 min. How we do it on Crusoe Cloud → crusoe.ai/resources/blog…

At 2am, no training node failure is an interesting engineering challenge. It's a page. A groggy SSH. A lost run. A ruined Tuesday. Somewhere we decided that was the cost of doing ML at scale. It's not. 🧵

Every token your model produces started as an electron. Not a metaphor, an engineering constraint. We asked five @CrusoeAI leaders what it actually takes to build an AI factory from power source to model output. Their answers converged on one idea: infrastructure designed around intelligence as the output changes everything underneath it. Read the full report: crusoe.ai/resources/lear…

Engineers building AI products: what's your biggest infrastructure headache right now? Our leaders have thoughts. → crusoe.ai/resources/lear…