Mr. Ånand

I've been using @nebiustf inference for the past few months across various projects. The open models like Qwen3-coder and Kimi K2 work well within IDEs like Cursor. We achieved 14,000 image generations for just $5 on one project. The pricing is competitive, with excellent support for numerous tools and agent frameworks like @langchain, @AgnoAgi, @crewAIInc, and others. It also offers one-click deployment, fine-tuning, and much more. I finally had some time to explore Nebius Token Factory. Watch the full video 👇

KitOps Community Call is live now. Get the latest updates regarding @Kit_Ops + quick bytes on Agent Guard → a zero-trust runtime to govern AI coding agents from @Jozu_AI If you’re coming to KubeCon EU, join the KitOps community and stay in touch with the team. Community Discord → discord.gg/eEucXzbg?event…

Most agent frameworks help you build agent logic. The harder part is everything after the demo: running agents in production. That’s why @AgentField_ai is interesting! It treats agents like real backend services, and adds a control plane around them. So infra is separate from agent code. Control plane handles: → orchestration → state management → identity → observability And you get infra features that are usually missing in agent frameworks: → async execution + long-running workflows → agent discovery + coordination via shared memory → cryptographic agent identities (DIDs) → verifiable execution records (auditability) → centralized orchestration + monitoring AgentField is more about operating agents as distributed systems. If agents are moving from prototypes → production, this control-plane layer feels inevitable.

More good news for the $NBIS token factory: Zixuan Li, the architect behind GLM, the smartest open model today, says the next iteration will be open. And it’s safe to assume it will be added to $NBIS' AI cloud. I wrote earlier today that this will likely be the year open models reach the next level of reasoning, potentially enough to solve century-old math riddles. And I wouldn’t be surprised if GLM 5.1 is the first to crack one. Who knows, maybe even on $NBIS' AI cloud. My earlier post: x.com/drtomslens/sta…

Congratulations, @InsForge_dev team! The repo is trending on GitHub🚀 If you’ve not used InsForge yet, it offers databases, auth, storage, and functions through a semantic layer that AI coding agents and AI code editors can understand, reason about, and operate end to end.

Composer 1.5 was good now looking forward to try Composer 2🔥

Most AI governance in regulated environments still runs on spreadsheets + email approvals. A model ships. An auditor asks: show me the exact training data, code version, and approval record for what’s running in prod right now. And suddenly you’re digging through S3 snapshots, Git commits, Slack threads, and half-updated docs. The issue isn’t documentation. It’s proof. Cryptographic proof that what’s deployed matches what was reviewed and approved. A direction that makes sense: turn compliance into a deployment gate, not a manual checklist. Core flow: → package model + code + prompts + datasets into one immutable ModelKit (content hash) → run AI-specific scans, attach signed attestations → enforce policies with OPA before push / pull / deploy If the artifact can’t prove: → what it contains → what it passed → who approved it …it doesn’t ship. What’s interesting is the @Jozu_AI scan layer goes beyond CVEs: → supply-chain checks on serialized models → prompt injection testing → adversarial robustness eval → LLM red teaming Each result is signed, so OPA evaluates verifiable metadata (not spreadsheet vibes). OPA policies can gate on: → risk level vs environment → required approvals → vuln thresholds → bias / robustness metrics Allow or deny + a reason. Decision logged → becomes part of the audit trail. Feels like SBOMs + signatures for containers… but applied end-to-end to ML artifacts. If regulators are moving toward continuous monitoring + tamper-proof logging, manual tracking won’t scale. Automated compliance gates feel like the realistic path. How are you handling AI governance today? Process discipline, or infra-enforced gates?

🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵

We’re hiring!

We released a new open-source MemWire, and @qdrant_engine was a natural fit to ship a zero-infra memory layer that anyone can self-host. GitHub repo: github.com/memoryoss/memw… Would love any feedback from the community! and stars⭐️

Jozu just launched Agent Guard → a zero-trust runtime for AI coding agents and MCP (Claude Code, etc). Most agent guardrails today run in user-space. So if the agent can access the same host processes/filesystem, a bypass is always on the table. @Jozu_AI Agent Guard 👉 hypervisor isolation + policy enforcement at runtime. What that means in practice: → agent runs in a hypervisor-isolated container → it only sees the workspace you explicitly share → host FS, processes, SSH keys, creds stay inaccessible → only supply-chain-verified artifacts get in → policies are tamper-evident + enforced on every action → policy server runs as PID 1 (agent can’t kill it) If your agent touches real credentials or prod-adjacent workflows, this is worth a look. I'll explore it in more details soon!

@sujee_dev @nebiustf Quite interesting 👀

🐍 Snakes at the Nebius booth @ Nvidia GTC '26🐍 …not that kind of snakes 😄 It’s the classic Snake game - but each snake is controlled by an LLM in real time. Powered by Nebius Token Factory @nebiustf 📍 Last day of the conference 🎮 Try different models and watch them compete live Stop by the @nebiusai booth 👀 #Snakegame #LLMGames

Glad to help @LiteLLM with the complete docs revamp! New doc is live - docs.litellm.ai

We’ve partnered with @nebiusai to bridge the gap between open-source models and production-scale deployment.🤝 Scaling architectures like MoE or sparse attention often pulls teams into weeks of “plumbing” — GPU orchestration, latency tuning, manual scheduling — work that adds zero core product value. We’re now putting optimized versions of models like DeepSeek, Qwen, and Llama directly into Token Factory. ⚙️ So teams don’t have to rebuild the whole stack every time. Focus on the product, not the infra. 👉Read the full case study: eigenai.com/blog/2026-03-1… #AI #LLMs #AIInfrastructure #Inference #Builders

Meet the new Stitch, your vibe design partner. Here are 5 major upgrades to help you create, iterate and collaborate: 🎨 AI-Native Canvas 🧠 Smarter Design Agent 🎙️ Voice ⚡️ Instant Prototypes 📐 Design Systems and DESIGN.md Rolling out now. Details and product walkthrough video in 🧵

We just dropped MiniMax-M2.5 by @MiniMax_AI 🔥 SOTA open-source agentic coder for production workflows & office tasks. Cheaper + more reliable agents 👇

@nebiustf @MiniMax_AI noice!