Build on The Grid
Agents, automations, art, video — running on community GPUs today. Pick one and ship.
Generate Images
Create AI art with Flux, SDXL, and Stable Diffusion. Thousands of models and workflows.
aipg.art →Run Agents & Workflows
Open-source LLMs powering your agents, automations, and AI workflows. Llama, Mistral, Qwen — no corporate worldview baked in.
aipg.chat →Build with the API
Drop-in for OpenAI and Anthropic. Same SDKs, real-time streaming, decentralized backend. Free daily quota.
API Docs →Earn with your GPU
Run an LLM, image, or video worker. Earn AIPG for every request you serve. Worker bond: 1,000 AIPG.
Run a Worker →Works with any OpenAI SDK. Three lines to get started:
AI shouldn't have a doorman.
$20 a month before your first line of code. A trial that expires. A card you don't have. Closed AI is full of velvet ropes. AIPG is built so the people who can pay fund the people who can't — and so the network can never be shut down by anyone, including us.
Free, and it stays free
Sign up, get a key, build. No trial counting down. No reminder email asking for your card. The free tier is the product — not a hook.
Pay it forward, automatically
When you pay for higher limits, you're not just buying compute — you're funding the free tier for someone shipping their first agent in a country where $20/mo isn't on the table.
Private. Unbiased. Yours.
Open models, no corporate worldview baked in by a policy team you've never met. On the confidential tier, even the operator running the GPU can't read your prompts.
Got a GPU? Put it to work.
The network runs on real hardware that real people plug in. If you have spare compute, you can earn AIPG by serving inference for everyone else.
AI Worker Node
Run an LLM, image, or video worker. Serve real requests for builders, agents, and apps — earn AIPG for every job you complete.
Validator Node
Sample inference jobs, verify the workers were honest, and put your stake behind that attestation. Real economic security — not vibes.
Got questions? Join our Discord and talk to the community.
How The Grid Works
Community GPUs run AI models. Smart routing sends your request to the best available worker. Workers earn AIPG for every job they complete.
GPU Workers
Gaming rigs to data centers — any GPU can join the network and start earning by running AI models.
Smart Routing
Requests automatically route to the best available worker based on model, load, and latency.
Token Rewards
Workers earn AIPG for every job completed. Each token represents real inference work done.
A network nobody can shut down
In DevelopmentEvery centralized AI you've used can be unplugged by a company, a regulator, or a Stripe ban. AIPG is moving entirely on-chain — smart contracts handle routing, payments, and verification. The team builds the protocol. The protocol runs itself. Once it's done, even we can't turn it off.
- ◆On-chain job anchoring with cryptographic proofs
- ◆Bonded validators verify inference results
- ◆Automatic payments via smart contracts
- ◆P2P network with no single point of failure
Privacy at the hardware level
Coming with BlackwellYour medical notes, your legal docs, your agent's API keys — they shouldn't be readable by the stranger running the GPU. NVIDIA Confidential Computing on Blackwell (B200/B300) encrypts your workload in GPU memory and proves cryptographically it ran in a secure enclave.
- ◆Hardware-isolated Trusted Execution Environment (TEE)
- ◆GPU memory encrypted end-to-end during inference
- ◆Cryptographic attestation proves genuine secure enclave
- ◆Run sensitive workloads on untrusted infrastructure
Open Source, Open Doors
All code, models, and workflows are public. No vendor lock-in. No black boxes. You can inspect everything, run your own instances, and build on top of what exists.
Open development means every improvement benefits everyone. Bug fixes, performance gains, and new features spread across the network immediately.
No Vendor Lock-In
Switch providers anytime. Run your own instance. Your data, your choice.
Faster Innovation
Every contribution benefits the whole network. Improvements compound across all users.
Community Owned
The network belongs to the people who build it and run it. No single entity controls AIPG.
Global Access
Same API whether you're in Buenos Aires or Bangalore. No waitlists, no geographic restrictions, no approval process.
Earn with Your GPU
GPU owners earn AIPG tokens for running inference jobs. Real work, real rewards. Your hardware serves actual users instead of sitting idle.
Useful Work
Every token represents real AI inference that served a real user. GPUs do productive work instead of arbitrary calculations.
Fair Launch
AI Power Grid launched with no VC, no presale, no pre-mine. Community-first from day one.
Dec 2023Building the Grid
Launched aipg.art for image generation, aipg.chat for LLM conversations, and the first GPU worker network. Public beta shaped by community feedback.
2024Base L2 Migration
Migrated to Base L2 with 150M fixed-supply AIPG token. Deployed StakingVault and core infrastructure contracts. Staking live with no lockup period.
Q1–Q2 2025Grid SDK & Dashboard
Released the Grid SDK for developers. Dashboard and API explorer went live. On-chain model registry for worker discovery.
Q3–Q4 2025Streaming API
✓ Live NowReal-time token streaming with OpenAI and Anthropic compatibility. Drop-in replacement — same SDKs, same code. Sub-second image generation with FLUX.
Q1 2026P2P Mode & Trusted Workers
Decentralized job distribution via libp2p gossipsub — no central coordinator required. Trusted worker bonding: stake AIPG for priority jobs and higher earnings.
Q2 2026Video & Token Factory
AI video generation from text prompts. ARC-20 Token Factory: launch tokens backed by real AI compute. Developer Launchpad with free inference for builders.
Q3 2026On-Chain Verification & NFTs
Worker registry and job escrow on Base L2. AI art NFTs with full generation parameters on-chain — not just a JPEG, the actual recipe. Dispute resolution via stake slashing.
Q4 2026Confidential AI
Private AI sessions using trusted execution environments. Verifiable, private inference where operators can't see your prompts or outputs.
2027