Built on Gonka — The AI Infrastructure Behind CloudMine
What does it actually mean to build AI infrastructure from scratch? Not just to spin up a few cloud VMs and call an API — but to design the economic layer, the compute layer, and the developer layer as an integrated system.
That's what MinglesAI did with CloudMine and Mingles Router. This article is about how those pieces work together, and why the architecture matters.
The Foundation: Gonka Network
Before CloudMine makes sense, you need to understand what it's sitting on.
The Gonka Network is a decentralized validator network powered by the GNK token. Like other Proof-of-Stake networks, it organizes activity into epochs — discrete time windows during which validators commit stake, earn rewards, and the network settles its accounts.
The core mechanics:
- Validator nodes lock GNK as stake to participate in the network
- Epochs are the settlement unit — each epoch, the network calculates rewards based on total weight and distributes GNK accordingly
- Network weight is the sum of all staked GNK across all participants at any given epoch
- GNK token is the native economic unit — used for staking, rewards, and billing
This is the base layer. It's a running, live network — at Epoch 186, the total network weight was 1,032,974 with rewards of approximately 295,687 GNK distributed across the epoch.
Layer Two: CloudMine
CloudMine is a GPU compute platform built on top of the Gonka network. The idea is simple: you invest GNK, CloudMine operates the infrastructure, and you earn GNK rewards every epoch.
You don't need to buy hardware, set up nodes, or monitor anything. CloudMine handles all of that. Your GNK is staked into an infrastructure contract — a position that earns a share of each epoch's reward pool proportional to the amount you've staked.
Each epoch, CloudMine's share of the total network weight determines how much of the reward pool flows to participants. Recent epochs show consistent returns:
| Epoch | Network Weight | Cloud Weight | Active Contracts | GNK Rewarded |
|---|---|---|---|---|
| 186 | 1,032,974 | 6,174 | 5 | 1,770.09 |
| 185 | 1,148,113 | 6,147 | 6 | 1,587.22 |
| 184 | 978,129 | 4,505 | 6 | 597.86 |
| 183 | 994,272 | 4,084 | 6 | 1,217.82 |
| 182 | 890,031 | 3,946 | 4 | 1,311.61 |
| 181 | 789,225 | 5,009 | 4 | 1,505.60 |
| 180 | 1,247,597 | 5,009 | 5 | 1,191.11 |
| 179 | 1,102,295 | 4,675 | 5 | 1,275.42 |
You can view live epoch data and your position's performance at cloudmine.mingles.ai/reports.
The Insurance Fund
One of the underappreciated mechanisms in CloudMine is the insurance fund.
GPU infrastructure is inherently variable — hardware fails, network partitions happen, utilization fluctuates. Beyond that, the primary risk investors should be aware of is that the network itself is still in active development: bugs appear frequently and updates roll out regularly. The insurance fund is a protocol-level reserve that provides epoch protection: if infrastructure underperforms in a given epoch due to circumstances outside anyone's control, the fund can cover the shortfall.
This matters for investors: it makes the yield profile more predictable. CloudMine positions behave more like infrastructure bonds than volatile trading positions. The insurance fund is funded by a portion of epoch rewards and operates transparently on-chain.
The Marketplace: Liquidity for Vesting Positions
GNK staking positions are typically subject to vesting periods — lock-up windows that prevent immediate withdrawal. This is standard for validator networks; it prevents rapid stake rotation that would destabilize the network.
The CloudMine marketplace solves the liquidity problem: holders of vesting positions can list them for sale, and buyers can acquire existing positions without the standard lock-up window. The price is determined by the market — a function of remaining vesting period, expected epoch yield, and buyer demand.
For investors, this is what turns a fixed staking position into a liquid asset class. The marketplace makes CloudMine positions tradeable without requiring any changes to the underlying protocol mechanics.
Layer Three: Mingles Router
The Mingles Router is where the network's compute capacity meets the real world.
Mingles Router (router.mingles.ai) is an OpenAI-compatible LLM inference API — the same interface developers already know, backed by the distributed GPU infrastructure that CloudMine powers.
GPU capacity staked into CloudMine infrastructure contracts is the same capacity serving inference requests through the Gateway. The network isn't just earning yield — it's running real AI workloads.
Web3 Billing
Mingles Router billing is token-based: you pay for inference in GNK, the same token used for staking and rewards. This keeps everything within the ecosystem.
Support for stablecoins and bank cards is coming in the future — making it accessible to anyone, regardless of their Web3 experience.
AI Agents on Gonka
One of the most exciting use cases for Mingles Router is AI agents. Agents need cheap, reliable, fast inference — and Gonka delivers all three, with an OpenAI-compatible interface that means zero migration effort.
Connecting an AI agent to the Gonka network is as simple as pointing it at the Gateway endpoint. No new SDKs, no special configuration — just swap the base URL and API key.
OpenClaw is one example of an agent already running on Gonka. OpenClaw is an AI development assistant that orchestrates code changes, manages issues, and coordinates development workflows — all powered by inference routed through Mingles Router. You can explore how agents connect at router.mingles.ai/agent.
As AI agents become a larger part of software infrastructure, having a decentralized, economically aligned inference layer becomes increasingly important. Gonka provides that foundation.
The Full Stack
Stepping back, here's what MinglesAI has built:
┌─────────────────────────────────────────────────────────┐
│ Agent & Developer Layer │
│ Mingles Router API — OpenAI-compatible LLM inference │
│ router.mingles.ai │
├─────────────────────────────────────────────────────────┤
│ Compute Layer │
│ CloudMine — GPU infrastructure contracts │
│ Epoch rewards · Insurance fund · Vesting marketplace │
│ cloudmine.mingles.ai │
├─────────────────────────────────────────────────────────┤
│ Network Layer │
│ Gonka Network — Validator epochs · GNK token │
│ Weight-based reward distribution │
└─────────────────────────────────────────────────────────┘
Each layer has a distinct function, and each depends on the layer below it:
- Gonka Network provides the economic foundation: decentralized validator consensus, epoch-based settlement, GNK token rewards
- CloudMine transforms that foundation into a simple investment: stake GNK, we mine, you earn — with insurance protection and a liquid secondary market
- Mingles Router exposes that compute capacity to developers and AI agents: OpenAI-compatible API, GNK-based billing, production-grade LLM inference
The result is a vertically integrated AI infrastructure stack — from validator economics through managed compute to developer API — operated by a single team.
Why This Architecture
The conventional approach to AI infrastructure is horizontal: pick a cloud provider for compute, pick an LLM provider for inference, wire them together with API calls and credit card billing. It works, but it creates dependencies at every layer — and the economic value flows outward to hyperscalers and LLM providers.
The Gonka/CloudMine architecture captures that value internally. Infrastructure contributors earn GNK from epoch rewards. Developers pay GNK for inference. The capital circulates within the ecosystem rather than leaking to external providers.
For investors, it's a straightforward proposition: stake GNK, earn GNK. CloudMine handles the rest.
Get Started
Both products are live and accessible today:
- CloudMine — Stake GNK, earn epoch rewards: cloudmine.mingles.ai
- Mingles Router — OpenAI-compatible LLM inference API: router.mingles.ai
For investors: review the epoch data at cloudmine.mingles.ai/reports, model your expected returns, and stake a CloudMine position.
For developers and AI agents: create an API key on Mingles Router and start routing inference through infrastructure you understand.
The network is running. The epochs are settling. The API is live.