0x3f2a9b…c417
Ed25519 / Merkle / libp2p
v2.0 · April 2026 · Living Document
Phase C · Protocol v2.0 · Zero blockchain required
A P2P Inference Protocol · 2026

Trustfromcryptography,not chains.

Pinaivu AI grounds every guarantee in Ed25519 signatures and SHA-256 Merkle proofs — not a coordinator, not a token. Settlement, storage and anchoring are pluggable. Swap a TOML value, not your stack.

Read Whitepaper v2.0
ED25519 IDENTITYSHA-256 MERKLE TREEGOSSIPSUB REPUTATIONAES-256-GCM SESSIONSX25519 CONTEXT KEYSSIGNED PROOF OF INFERENCESETTLEMENT-AGNOSTIC ESCROWLIBP2P · QUIC · NOISEIPFS · WALRUS · LOCALFREE · RECEIPT · CHANNEL · SUI · EVMSTANDARD · PRIVATE · FRAGMENTED · MAXIMUMOFFLINE VERIFIABLE
Scroll
0/5
Guarantees met G1–G5
0
Blockchains required Optional
0
Stack layers Swappable
0-bit
Ed25519 security RFC 8032
§ 001 Thesis · Drafted for the open network · v2.0
Abstract · Line 01Self-sufficient

Every prior inference marketplace grounds trust in a coordinator or a specific chain. Pinaivu AI takes a third path: trust is grounded exclusively in cryptography — Ed25519 identity, SHA-256 Merkle proofs, AES-256-GCM sessions. Any chain becomes an optional anchor on a system that already works.

Offline verifiableNo coordinatorChain-optional
Primitive · 01
Proof of Inference
A signed execution receipt verifiable offline with only the producing node's public key.
π = (req, model, tᵢ, tₒ, Δ, H_in, H_out, pk, σ)
# verify π offline — no network, no chain
let msg = canonical(π)
let vk = VerifyingKey::from_bytes(π.pk_N)
assert EdDSA::verify(vk, msg, π.σ)
# O(1) — constant time
G1
Session privacy
Client-held KX25519 DH
G2
Node accountability
Ed25519 σMerkle π
Zero blockchain required
G3
Settlement neutrality
free · receipt · channel · sui · evm
G4
Storage neutrality
local · ipfs · walrus
5/5
G5
Permissionless participation
libp2p PeerId = pk_N
Reputation
score(N) = α·ṡ + β·ℓ
α=0.6β=0.4L_max=5s
Gossip
600s · broadcast root
/pinaivu/reputation/1.0.0
002 · The Failure Mode

Cloud AI bakes in three consequences
that aren't technical requirements.

For every turn (P, C, R), today's provider observes all three, sets price ρ unilaterally, and revokes access at will. None of this is forced by the maths — only by the architecture.

01 — Context exposureG1

Provider sees (P, C, R)

Every prompt, every accumulated context, every response flows through one party. Pinaivu AI keeps the full session `C` encrypted under a client-held key `K`; the GPU node sees only the decrypted context window for the current turn.

02 — Chain dependenceG3

One token, one ecosystem

Bittensor collapses without TAO. Every prior decentralised inference system grounds trust in a specific chain, token and validator set. Pinaivu AI's trust model is self-sufficient; any chain is an optional settlement adapter selected in a TOML file.

03 — Unverifiable workG2

No receipt, no recourse

Batch marketplaces (io.net, Akash) and routers (Fortytwo) can't prove node N ran job J at the claimed parameters. Pinaivu AI ships every response with a self-verifiable ProofOfInference — Ed25519-signed, offline checkable, binding on (model, tokens, Δ, H_in, H_out).

003 · Six Layers

Every layer is independently replaceable.

Layers interact only through trait interfaces. Layer 0 (Crypto) has no external deps. Every layer above it may use external infra, but none is required.

L · 06 · Application

OpenAI-compatible surface

TypeScript SDK, drop-in HTTP API, Web UI. Change the base URL; keep your code. Streaming, sessions and proof retrieval are native.

TS SDK · HTTP · Web UI
L · 05 · Session

E2E encrypted memory

Full history `C` is AES-256-GCM encrypted under a client-held `K`. The GPU node decrypts only the active context window — never `C`, never `K`.

AES-GCM · X25519 · Portable
L · 04 · Reputation

Merkle tree, gossiped

Every node keeps a Merkle tree of its signed proofs. The root is broadcast over libp2p gossipsub every 10 min. Chain anchoring is optional.

SHA-256 · Gossipsub · O(log n)
L · 03 · Marketplace

200ms sealed-bid auction

Client broadcasts request; nodes pass six cheap-to-expensive checks and submit a bid. Composite score (0.4×price + 0.3×latency + 0.3×rep) picks the winner.

libp2p · Sealed-bid · First-price
L · 02 · Settlement

Pluggable escrow

Five adapters: free, signed-receipt, off-chain channel, Sui, EVM. Pick in TOML; same binary. Payment channels amortise gas 50× over 100 requests.

free · receipt · channel · sui · evm
L · 01 · Storage

Content-addressed, agnostic

Three-method interface: put/get/delete. Local, IPFS, Walrus, Memory — same protocol. SHA-256 IDs mean put(b)=put(b) deduplicates for free.

local · ipfs · walrus
004 · Request Flow

From prompt to proof, in under a second.

Four stages. Each one cryptographically verifiable — from the sealed-bid auction through Ed25519-signed proof delivery.

1
Step 01 · ~5ms

Broadcast

Client broadcasts an InferenceRequest on the gossipsub topic for the required model, carrying model ID, budget, and privacy level — not the context (that stays client-side until a winner is chosen).

2
Step 02 · 200ms

Sealed-bid Auction

GPU nodes pass six checks (model, capacity, queue, budget, privacy, throttle) and submit bids. Client picks winner by composite score: 0.4×price + 0.3×latency + 0.3×reputation.

3
Step 03 · ~620ms

Inference

Client encrypts the context window W for the winning node via X25519 DH and sends it directly to that node's API. Node decrypts W in RAM, runs inference, streams tokens back, then zeroes W.

4
Step 04 · ~20ms

Proof + Settle

Node signs ProofOfInference π binding (model, tokens, Δ, H_in, H_out) with Ed25519. π is appended to the node's Merkle tree. Settlement adapter executes and ships π to the client.

0ms
Total · end to end
42tok/s
Throughput · 70B model
0.0003 PEER
Cost · 256 tokens
005 · Model Catalog

Run the models you want. Not the ones they allow.

Every open-weight checkpoint that fits in VRAM. Pre-cached for the popular ones, on-demand for the rest.

84 models live · 2,847 variants
LLMTextFP16 · INT8 · INT4

Llama 3.1 · 405B

Meta · Open weights · Released Jul 2024

The largest open LLM running on the network. Sharded across 16 consumer GPUs via tensor parallel. Competitive with GPT-4 on most benchmarks at a fraction of the cost.

Parameters
405B
Context
128K tokens
Throughput
42 tok/s
Cost / 1K
$0.003
peer-cli · llama-3.1-405b
VisionDiffusion1024²

FLUX.1 · Pro

Black Forest Labs · Open weights · Aug 2024

State-of-the-art text-to-image at 1024² native resolution. Runs on a single consumer GPU. 4-step Turbo variant generates in under 1 second per image.

Resolution
1024×1024
Steps
4 (turbo)
Latency
2.1s
Cost / img
$0.004
peer-cli · flux-1-pro
AudioSTTStreaming

Whisper · Large v3

OpenAI · Open weights · MIT license

99-language speech-to-text with automatic language detection. Runs 52× realtime on an RTX 3090. Native WebSocket streaming for voice applications.

Languages
99
Speed
52× realtime
TTFT
<300ms
Cost / min
$0.001
peer-cli · whisper-v3-large
006 · Comparison

Against the incumbents.

Every prior system either lacks G2 (no verifiable accountability) or sacrifices G3/G4 (hard-coded chain and storage). Pinaivu AI is the first to satisfy all five guarantees simultaneously.

Property
Pinaivu AI
Bittensor
QVAC
io.net
Fortytwo
G1 — Session privacy
AES-256-GCM
Validators see all
Not addressed
N/A · batch only
Centralised
G2 — Node accountability
Ed25519 + Merkle
Partial · validators
No receipts
No receipts
No receipts
G3 — Settlement neutrality
5 adapters
TAO only
No payment
IO token
N/A · centralised
G5 — Permissionless
PeerId = pk_N
KYC required
Centralised
Persistent sessions
E2E encrypted
Streaming responses
Native WebSocket
007 · Stack

Built on proven primitives.

No reinvention for its own sake. Every layer is a battle-tested open-source component, assembled specifically for GPU compute coordination.

libp2p Transport

T · 01

TCP + QUIC dual-stack with Noise authenticated encryption and Yamux stream multiplexing. AutoNAT traversal means any home node can participate without port-forwarding.

TCPQUICNoiseYamux

Kademlia DHT + Gossipsub

T · 02

Kademlia DHT for peer routing and mDNS for local discovery. Five gossipsub topics carry inference requests, bids, announcements and Merkle root broadcasts.

KademliamDNSGossipsub5 topics

Ed25519 Identity

T · 03

Every node is an Ed25519 keypair. The libp2p PeerId is derived from pk_N — no separate account or wallet needed. 128-bit security per RFC 8032.

Ed25519RFC 8032128-bit security

ProofOfInference

T · 04

A signed execution receipt bound to (model, tokens, latency, H_in, H_out). Verifiable offline with only the node's public key. Constant-time O(1) verification, no network call.

Ed25519 σSHA-256 H_in/H_outOffline

AES-256-GCM Sessions

T · 05

Session context encrypted under a client-held key K derived from X25519 DH. The GPU node never sees K — only the current-turn context window, zeroed from RAM after inference.

AES-256-GCMX2551996-bit nonce

Settlement Adapters

T · 06

Five adapters behind one interface: free, signed-receipt, off-chain payment channel, Sui (Phase D), EVM (Phase E). All selected by a single TOML key — same binary, zero code changes.

freereceiptchannelsuievm
008 · Fleet

The GPUs behind the mesh.

A live breakdown of the hardware running inference right now. Consumer cards dominate the network — by design.

RTX 4090
24GB · 82.6 TFLOPS
Network share68%
RTX 3090
24GB · 35.6 TFLOPS
Network share18%
A100 · 80GB
80GB HBM2e · 312 TFLOPS
Network share9%
Other
4080 · 4070 · M-series · more
Network share5%
009 · Timeline

From testnet to full mesh.

Four phases. Shipping cadence tied to node-count milestones, not marketing dates.

Live
Phase C · April 2026
Cryptographic Core
  • Ed25519 identity + ProofOfInference
  • Merkle reputation tree + gossip
  • Free + signed-receipt settlement
  • Local + IPFS + Walrus storage
Queued
Phase D · H2 2026
Sui Settlement
  • Move escrow smart contract
  • SuiSettlement adapter live
  • On-chain proof verification
  • Reputation anchoring on Sui
Queued
Phase E · H1 2027
EVM Settlement
  • Solidity escrow contract · Base L2
  • EvmSettlement adapter live
  • Multi-chain settlement matrix
  • TOML-selectable chains
Queued
Phase F · H2 2027
On-Chain Channels
  • Payment channels — on-chain close
  • 50× gas amortisation at 100 req/session
  • Full gossip protocol live
  • Governance parameterisation
— 010 · Start Here

Be first on the network.
Join the waitlist.

No credit card. No token. No permission. Phase C is live — Ed25519 identity, Merkle reputation and signed-receipt settlement work today, with zero blockchain required.

Read Whitepaper