Introduction to GridAI

GridAI represents a paradigm shift in industrial-scale artificial intelligence. By aggregating decentralized consumer hardware—specifically arrays of NVIDIA GPUs—GridAI constructs a formidable, highly scalable computing cluster that rivals traditional hyperscalers at a fraction of the cost.

Quick Start

Getting started with GridAI is designed to be frictionless. As a user, you do not need to configure heavy local environments or manage complex dependencies.

1. Access the Dashboard

Navigate to the GridAI Studio (deployed on Vercel) and authenticate using your secure credentials.

2. Choose from 8 Industrial Applications

Select from our massive suite of 8 native AI engines: Image Generation (SDXL), Language Models (Llama3), Text-to-Speech (F5-TTS), Video Dubbing, AI Avatars (Wav2Lip), Advanced Lipsync (SadTalker), Video Generation (SVD), and Headless Rendering (FFMPEG/Playwright).

3. Execute on the Grid

Your task is dynamically assigned to the most optimal decentralized node via the Mempool ledger. Results are returned when execution completes.

Architecture: The Decentralized Mempool

GridAI operates on a highly resilient, fully decentralized Mempool Architecture. We have securely bridged Vercel (Frontend), Render (Cloud Orchestrator), and Local GPUs into a unified, zero-latency network.

Vercel & Render Integration

The ultra-fast Next.js frontend is hosted on Vercel, providing instant interaction globally. When an AI job is dispatched, it hits the Cloud Orchestrator (Node.js engine hosted on Render), which natively writes it to the MongoDB Mempool Ledger.

Zero-Downtime Tunnel Registry

Local GPUs securely expose their microservices via localhost.run tunnels. The start-gpu-tunnels.js script actively monitors these 8 connections and seamlessly re-registers their dynamic public URLs to the Render Cloud Orchestrator, guaranteeing zero downtime.

The Local IPv4 Loopback Engine

To bypass the 1MB payload limits of free SSH tunnels when handling heavy Video files, the Local Orchestrator forces all 8 nodes to sync using the raw 127.0.0.1 IPv4 Loopback. The Local Mempool Bridge downloads tasks from the cloud, feeds massive payloads to local GPUs at maximum PCIe speeds, and pushes the final product back to the cloud database asynchronously.

State-of-the-Art Voice Cloning: F5-TTS

GridAI recently upgraded its core neural voice engine from Coqui XTTS-v2 to F5-TTS, the absolute bleeding-edge open-source zero-shot voice cloning model.

Why F5-TTS?

Legacy TTS architectures like XTTS natively suffer from the "robotic kitten" degradation effect when exposed to complex reference audio. F5-TTS solves this fundamentally by thriving on larger context limits. Instead of a strict 6-second constraint, the new engine seamlessly parses up to 15 seconds of raw, uncompressed reference audio. This allows it to natively capture the deep nuances, breath, and emotional timbre of the speaker flawlessly, while completely bypassing any artificial metallic degradation.

Industrial Reliability (Zero Downtime)

GridAI implements advanced fault-tolerance mechanisms specifically designed for enterprise environments. Our system guarantees near 100% uptime, even if individual node hardware suffers catastrophic failure mid-generation.

🫀 The Industrial Heartbeat System

Instead of relying on arbitrary static timeouts, GridAI uses active telemetry. When a Decentralized Node claims a job, it spawns a background Heartbeat Thread. While the heavy AI model is rendering, this thread pings the Mempool every 15 seconds to confirm the node is still alive.

🚑 The Dead-Man's Switch

The network constantly monitors the pulse of all active jobs. If a node loses power, crashes, or loses internet, its heartbeat immediately stops. If any processing job fails to receive a ping for 45 seconds, the Dead-Man's Switch triggers. The orphaned job is instantly rescued and placed back into the pending queue, where a surviving node will immediately pick it up and finish the task.

Contribute Compute: Running a Worker Node

GridAI relies on a decentralized network of GPUs. Because we use a Mempool architecture, hosting all 8 nodes requires no inbound open ports, no port-forwarding, and no router configuration.

The Unified Node Launcher

To activate your hardware and instantly deploy all 8 neural engines, you just need to run the master script. It automatically launches the LLM, Image, Video, Audio, Avatar, Dubbing, and Headless rendering servers into a clean, tabbed Windows Terminal interface.

Windows Users

Simply double-click the included batch file in the root directory:

START_ALL_NODES.bat

Tunnel Activation

The batch file will automatically trigger start-gpu-tunnels.js, syncing your local nodes with the Render Cloud Orchestrator and preparing them to execute jobs directly from Vercel.

Industrial Frontend Sync Engine

GridAI Studio is equipped with an advanced Auto-Resume Sync Engine designed for long-running GPU tasks.

Stateless Auto-Resume

Traditional Web UIs lose context of background server tasks when you refresh the page or close the tab. GridAI's Frontend Sync Engine completely solves this by persisting active decentralized Job IDs into local browser storage.

If you execute a 15-minute Video Dubbing job and accidentally close the tab or refresh your browser, you will not lose your progress. The moment the dashboard mounts, the Sync Engine automatically reads the active ledger, shifts the UI back into "Generating" mode, and seamlessly reconnects to the Mempool's active telemetry feed to resume your progress bar exactly where it left off.

Zero-Trust Privacy

Enterprise Security Standard

Proprietary data never leaves the decentralized network. GridAI processes data securely on local silicon or verified nodes.

GridAI operates on a strict Zero-Trust model. By moving completely away from cloud APIs and Master Orchestrators, your data is executed purely in a P2P micro-environment.

Decentralized Execution: All Heavy processing (SD, LLM, Video) is done securely by anonymous worker nodes.
No Ingress Attack Vectors: Because the network relies on outbound Mempool polling, there are zero open ports exposed to the public internet.
Ephemeral Processing: Once an AI task is completed and written to the database, data is immediately flushed from the node's local memory.

Hybrid Web2/Web3 Billing

GridAI implements a massive Hybrid Gateway integrating traditional Fiat processing (Stripe) and Decentralized Web3 Escrow (Polygon Testnet).

💵 US Dollars

Users can instantly purchase 5,000 Compute Credits for $5.00 USD via Stripe and PayPal. Backend webhooks securely provision credits into the centralized GridAI database.

⛓️ USDC (Polygon)

Crypto users can purchase 5,000 Credits for 5 USDC. The frontend connects to MetaMask, prompting an ultra-fast (<2s) transaction on the Polygon network.

Roadmap: True Decentralized Ledger (Phase 2)

Now that GridAI has successfully migrated to the Mempool Architecture, our next milestone is replacing the MongoDB backend with a purely distributed ledger.

1. libp2p Gossip Protocol
Instead of polling a central MongoDB cluster, active GPU nodes will gossip their hardware availability and completed blocks directly to the decentralized network using the IPFS network stack.
2. Proof of Compute & Staking
Node operators must stake the native **GRIDAI Coin** to participate in the network. Slashing mechanics will programmatically penalize malicious actors attempting to fake compute results.
3. Tokenomic Deflation (Buy & Burn)
To secure the network's value, a percentage of all USDC revenue routed through the decentralized payment gateway will be automatically utilized to buy back and burn GRIDAI coins from the DEX.