{
  "uid": "cogitave.learn.docs.what-is-yuva",
  "kind": "doc",
  "href": "/docs/what-is-yuva/",
  "title": "What Yuva is",
  "summary": "Yuva is a from-scratch, agent-native sovereign unikernel in no_std Rust, with formally verified leaves and a boot that fails closed. What it is for, how it is verified, and exactly which parts are not live yet.",
  "type": "explanation",
  "products": [
    "yuva"
  ],
  "roles": [
    "developer",
    "solution-architect",
    "platform-engineer"
  ],
  "levels": [
    "intermediate"
  ],
  "subjects": [],
  "headings": [
    "Why build a kernel at all",
    "How it is built",
    "Current status - read this before planning against it",
    "What this means for you",
    "Next steps"
  ],
  "source": "\n# What Yuva is\n\n**Yuva is a from-scratch, agent-native sovereign unikernel** - a kernel and\nmicro-VMM written in `no_std` Rust, built to run agent workloads on hardware it\nfully controls rather than as a process inside somebody else's operating system.\n\nIt is a different layer from [Namzu](what-is-namzu.md), and the two are not a\nsingle stack. Namzu is a TypeScript kernel you install from npm. Yuva is the\nfloor underneath: it boots on bare metal.\n\n## Why build a kernel at all\n\nAn agent that can be observed, paused, or read by the platform hosting it is not\nsovereign. Every guarantee you make about an agent's isolation, its memory, and\nits audit trail is only as strong as the layer enforcing it. Yuva exists so that\nlayer is one Cogitave wrote, can verify, and can attest to - not one it rents.\n\n## How it is built\n\n- **`no_std` Rust, zero `unsafe` above one audited HAL crate.** Unsafe code is\n  confined to a single hardware-abstraction crate that is reviewed as a unit;\n  everything above it is safe Rust.\n- **Formally verified leaves.** Selected components are proved with Kani, and the\n  harness counts are pinned in CI - the proof set cannot silently shrink.\n- **Boots two architectures on every push.** `x86_64` and `aarch64`, on every\n  commit.\n- **Fails closed.** The boot does not report success unless the full cumulative\n  self-test marker chain prints over serial. A missing marker is a failed boot,\n  not a warning.\n\nThat last property is the one worth internalizing: correctness is enforced by\nconstruction and checked on every push, so a regression cannot ship quietly.\n\n## Current status - read this before planning against it\n\nYuva's own README states its limits in uppercase, and this page repeats them\nrather than softening them:\n\n> [!IMPORTANT]\n> **Live inference is MOCK. Learning is DORMANT** (gate-not-met). The kernel\n> reports its own status in honesty tokens over serial rather than implying a\n> capability it does not have.\n\nSo Yuva today is a verified boot, a verified kernel substrate, and an honest\nreport of what is not switched on. It is **not** a runtime you can deploy an\nagent onto today, and no page here will tell you otherwise. If you need to run an\nagent now, that is [Namzu](what-is-namzu.md) on Node.\n\n## What this means for you\n\n| If you want to | Use |\n|---|---|\n| Run an agent today | [Namzu](what-is-namzu.md) - `@namzu/sdk` on Node 20+ |\n| Understand the sovereignty argument | This page |\n| Track when Yuva can host workloads | The honesty tokens are the signal; they change when the gate is met |\n\n## Next steps\n\n- [What Namzu is](what-is-namzu.md) - the layer above, and the one you can use now.\n- [Agent identity and capabilities](agent-identity-and-capabilities.md) - the\n  authorisation model both layers enforce.\n",
  "lastReviewed": "2026-07-26",
  "order": 2
}