> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pinaivu.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> Pinaivu — private, peer-to-peer LLM inference over a decentralized GPU network

## What is Pinaivu

Pinaivu routes private LLM inference requests to a **decentralized network
of independent GPU node operators**. A **coordinator** brokers each
request: it runs an open auction among bidding nodes over libp2p,
dispatches the winner, and signs an auditable routing receipt once the job
completes.

The hard problem this project solves isn't routing requests to GPUs — it's
doing that broker job **without asking anyone to trust the company running
it**. A real-time auction needs *someone* to rank bids and pick a winner in
milliseconds, which is inherently a coordination role, and that role can't
itself be a leaderless protocol without sacrificing latency. So Pinaivu
separates concerns cleanly instead of pretending the coordinator doesn't
exist:

* **The compute layer is genuinely decentralized.** Any operator can join
  the open libp2p mesh and bid — no permission from Pinaivu required.
* **The coordinator is a single off-chain component, but it's verifiable,
  not trusted.** It runs inside an attested Nitro Enclave (the Nautilus
  pattern) and every artefact it signs is checkable against an on-chain
  record it cannot quietly alter.

<CardGroup cols={2}>
  <Card title="Nautilus / Nitro Enclave" icon="microchip" href="/architecture/decentralization">
    The coordinator and chat-relayer are off-chain components that run
    inside AWS Nitro Enclaves, so their code — not just their word — is
    cryptographically attested.
  </Card>

  <Card title="Sui contracts" icon="link" href="/architecture/onchain">
    On-chain Move modules verify the attested enclave key and gate every
    payout — the mechanism that makes the off-chain components verifiable.
  </Card>

  <Card title="Walrus" icon="database" href="/architecture/decentralization">
    Routing receipts and chat history live on independent, content-addressed
    storage, not a private database anyone could quietly edit.
  </Card>

  <Card title="Decentralized GPU network" icon="circle-nodes" href="/architecture/overview">
    Inference itself runs on independent operators' hardware over an open
    libp2p mesh — the coordinator brokers, it never computes.
  </Card>
</CardGroup>

## Where to go next

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Make your first request against the API
  </Card>

  <Card title="Architecture" icon="diagram-project" href="/architecture/overview">
    How the coordinator, nodes, and explorer fit together
  </Card>

  <Card title="Decentralization & verifiability model" icon="shield-halved" href="/architecture/decentralization">
    What's decentralized (the GPU network), what's verifiable off-chain
    (the Nautilus components), and what's on-chain (the contracts that
    tie them together)
  </Card>

  <Card title="Why Sui" icon="circle-question" href="/architecture/why-sui">
    Move, Nautilus, Walrus, and gasless settlement — why the on-chain layer
    had to be Sui specifically
  </Card>

  <Card title="API Reference" icon="code" href="/api-reference/introduction">
    The OpenAI-compatible developer surface
  </Card>
</CardGroup>