Abakos Whitepaper

Abakos: A PoS Compute Network for Maximum Hardware Income

Whitepaper v0.2, draft. Public-facing. Figures and parameters are illustrative starting assumptions, not final.

Hardware that stays fully used.

Abstract

Abakos is a Proof-of-Stake Layer-1 for decentralized compute, forked from open Akash-style marketplace modules (Cosmos SDK + CometBFT) onto its own chain so ABA can be the settlement and staking asset. The product thesis is maximum income from every machine. It runs the product suite Akash proved, the Console for deploys, Abakos Chat, and an OpenAI-compatible API, on its own chain. Buyers deploy resource bundles (CPU + RAM + disk, optional GPU) and optional add-ons (persistent storage, IP lease) through the Console, paid from an ABA wallet. When GPU or CPU capacity would sit idle, a Provider Agent mines the most profitable coin, auto-converts the proceeds into ABA, and pays the host. Providers are not paid by minting ABA for compute. Earlier Proof-of-Useful-Work / btcd research is archived and is not the live product claim.

1. Background & motivation

AI and cloud workloads need cheap, available compute. Hyperscalers still control most data-center GPU capacity. Decentralized rental networks proved that self-serve container deployment can work (Akash, Vast.ai and similar), but they leave a structural hole: idle GPU and CPU hours usually earn nothing. Mining-only DePIN designs often mint tokens to hosts whether or not anyone bought capacity, which decouples provider pay from real demand.

Abakos starts from Akash-like rental rails, then adds an Agent that keeps idle compute earning by mining the most profitable coin and auto-converting it into ABA, plus AI-native demand surfaces (Chat and API). Settlement at MVP is ABA wallet only. Fiat credits (card to market-buy ABA) may come later as an onramp, not as a substitute for ABA demand.

2. Design principle: utilization first, no provider inflation

if demand for slice/add-on:
    earn ABA from buyer escrow
else if GPU or CPU free:
    mine best coin -> auto-convert -> ABA to host (12% cut; 88% to host)
else:
    storage / RAM wait for rental (no mining path)

• Compute hosts are paid from real money paths (buyer ABA or mining

proceeds auto-converted to ABA), not from minting ABA for idle work.

• Chain security is paid from **protocol fees plus the staker share of the

mining and Chat cuts - ABA has zero inflation** (fixed 10B supply), not a minting subsidy.

• Buyers never rent a naked GPU. A deployment is always a bundle.

3. Architecture

3.1 Consensus & chain

• Own PoS app-chain, CosmWasm-capable Cosmos stack, forked so ABA captures

fees, genesis allocation and validator economics.

• Technical base: open Akash marketplace modules (Apache-2.0 lineage) adapted

to Abakos parameters and settlement.

• Native EVM (cosmos/evm, EIP-155 chain id 9721): Ethereum JSON-RPC and

MetaMask alongside the Cosmos side, so Solidity contracts run natively.

• Zero-fee (gasless): transactions cost nothing on both the Cosmos and EVM

sides (no_base_fee, min_gas_price = 0, eth_gasPrice = 0); spam is bounded by consensus block.max_gas and a capped mempool, not by price.

• On-chain ABA/USDC DEX: a Uniswap-v2 fork (0.30% swap fee to LPs) is the

buyback venue and price oracle. Stablecoin standard is USDC (Noble, over IBC).

• Earlier PoUW / GEMM / reward-split work remains research archive only.

3.2 Console & marketplace

Two entry points into the same settlement rails:

• Templates / hourly bundles: one-click or browse; auto-match cheapest

qualified provider or pick a listing.

• Funded batch jobs: used by the Developer API and Abakos Chat; ABA

locked in escrow until delivery.

What buyers pay for (not all "free in the container"):

Providers list once and stay active. Matching assigns funded work automatically. That is required for Chat and streaming API UX.

3.3 Provider Agent

• Serves paid Console jobs and rentals first.
• When GPU/CPU would be idle: mine the most profitable coin, auto-convert,

pay host in ABA.

• Protocol cut on buyback conversion: 12% (88% to host; 4% to stakers,

4% to treasury, 4% burn). Conversion is automatic; hosts always earn the network asset.

• Payout by verified shares: miners connect to the Abakos stratum proxy with

their ABA address as the login; the proxy relays to the upstream pool (unMineable, auto-exchanged and bridged to USDC) and counts difficulty-weighted accepted shares per address. The Agent pays each host proportional to those verified shares, so self-reported hashrate cannot inflate a payout.

• Provider Dashboard shows utilization, rentals, add-on earnings, mining

buyback, connected providers, verified shares, uptime and reputation.

3.4 Model registry & AI workloads

Open-weight models only for network-served inference (Llama, Qwen, DeepSeek and similar). Closed frontier APIs are never falsely claimed. Large models stay inside one provider's multi-GPU box; no weight sharding across the public internet in v1.

3.5 Current implementation status (Public sandbox live)

Radical transparency is part of the design. Live status: status.abakos.ai.

• Live today (public sandbox): PoS chain + EVM (id 9721), zero-fee

transactions, web wallet, public explorer, on-chain ABA/USDC DEX with real Circle USDC (Noble, over IBC) and in-app bridge deposits from Polygon and other chains, real mining payouts (mined value -> USDC -> on-chain ABA buyback), and the Provider Agent + Dashboard with payout by verified shares. Single-operator for now.

• In design / not live: Console (bundles, add-ons, on-chain escrow),

Developer API, Chat, external validators (decentralization), fiat onramp, mainnet.

• Nothing above is a ship-date promise.

4. Abakos Chat & Developer API

• Developer API (Phase 3): OpenAI-compatible batch embeddings and

completions that create Console jobs settled in ABA. Open-weight models, the same idea as a managed inference API like AkashML.

• Chat + streaming (Phase 4): consumer demand on the same rails. Chat

may later hide crypto on the surface; MVP settlement remains ABA. Product markup: +12% (4% stakers, 4% treasury, 4% burn); provider net matches the Console, then the usual 3% fee applies.

5. Economics & token (ABA)

• Genesis supply: 10,000,000,000 ABA (fixed), zero inflation. No

provider mining-emission allocation.

• Illustrative allocation: Liquidity 32%, Treasury 18%, Ecosystem 15%,

Reserve 15%, Team 12% (1-year cliff, 3-year linear), Community 8%.

• Illustrative DEX start: ~$0.002 (~$20M FDV), staged liquidity seeding.
• Transaction fees: zero (gasless L1; eth_gasPrice = 0). The figures

below are the protocol revenue share on real activity, not gas.

• Protocol revenue:
• Console / API / CPU / storage settlement: 3% = 1% stakers + 1% burn +

1% treasury.

• Idle mining buyback cut: 12% (88% host, 4% stakers, 4% treasury, 4% burn).
• DEX swap fee: 0.30% - 0.25% to liquidity providers, 0.05% protocol

(1/3 stakers, 1/3 treasury, 1/3 burn).

• Chat / API product markup: +12% = 4% stakers + 4% treasury + 4% burn,

then standard settlement fee.

• Stablecoin standard: USDC (Noble, over IBC) for pool payouts and the ABA/USDC DEX

(0.30% swap fee to LPs).

• Burn: a burn slice on every revenue source (1% of Console/Marketplace

settlement; 4% of the idle-mining, Chat and API cuts) permanently removes ABA from supply. Zero inflation plus these burns makes ABA net deflationary.

• MVP payments: ABA wallet. Fiat to ABA onramp post-launch.
• Fundraising preference: compute vouchers / grants / strategic capital

against fixed-supply design, not a public token-sale page.

6. Security & trust

• Novel surface: marketplace escrow, Agent buyback pipeline, matching and

Console tooling. Base Cosmos/Akash lineage reduces reinventing consensus from scratch, but Abakos-specific modules need audit before mainnet.

• Launch gate: public sandbox (live) → external audit → mainnet + liquidity.
• Honesty policy: status badges distinguish live vs planned; no claim that

PoUW reward-split is the product.

7. Go-to-market (summary)

1. Provider Agent + Dashboard (supply and utilization story).
2. Console templates (demand that fills hardware).
3. API batch, then Chat/streaming.
4. Fiat onramp when UX and compliance justify it.

8. Roadmap

Phase numbers match status.abakos.ai.

• Phase 0 (done): Architecture pivot.
• Phase 1 (live): Public sandbox - PoS chain + EVM, zero-fee, wallet, explorer, ABA/USDC DEX, Provider Agent + Dashboard.
• Phase 2: Console, bundles, add-ons, ABA escrow, idle buyback.
• Phase 3: Developer API (batch).
• Phase 4: Chat + streaming API.
• Launch: Audit, liquidity, mainnet.
• Post-launch: Fiat to ABA, broader VPS/storage products.

9. Risks

Bootstrap (supply and demand must both show up); oracle and conversion risk on idle buyback; ABA price volatility for hosts and buyers in a wallet-first MVP; latency for decentralized inference (hybrid capacity may be labeled explicitly in early Chat); competition from Akash, Vast.ai, io.net and others (differentiate on utilization Agent + ABA economics + Chat/API); legal classification of ABA (counsel before any public sale). Internal numbered docs under docs/ may still contain older PoUW wording until rewritten; prefer this whitepaper and the website for the current thesis.

10. References

• Akash Network docs (SDL, providers, IP leases, persistent storage):

https://akash.network/docs/

• Akash AEP-76 (Burn-Mint Equilibrium / credit card → AKT demand design):

https://akash.network/roadmap/aep-76/

• Pearl / PoUW background (historical contrast, not Abakos product claim):

https://pearlresearch.ai/

This document is for information only and is not financial advice or an offer of securities. ABA parameters are drafts subject to legal and audit review.