Decentralized compute / Open source models / Bitcoin

AI agents
for everyone

A peer-to-peer network that turns the world's hardware into AI infrastructure. No censorship. No permissions. No middlemen.

see how it works

Centralized AI doesn't scale for everyone

Today you depend on a handful of Silicon Valley companies to access artificial intelligence. That comes at a cost.

Expensive

Obscene margins on cloud GPU time. $20/month for a rate limit that's never enough.

Censored

Providers decide what you can and can't do with your own tool.

Dependent

If they cut your access, you lose everything. Your work, your data, your flow.

Inequitable

All infrastructure is in the US. The rest of the world pays for latency and inflated prices.

Regionalized P2P network

Geographically distributed compute nodes. Each region processes its own requests. Your data doesn't cross borders.

Open the app

Web, mobile, desktop. Looks like any AI chat. You don't know or care that it's P2P.

Your request hits the router

A smart DNS (the only centralized piece) knows available nodes, their capacity, and their location.

A nearby node is assigned

The router finds the nearest node with available GPU. Tokyo, Berlin, São Paulo — wherever there's capacity.

Local inference, direct streaming

The node runs the model (Llama, Mistral) locally. Tokens stream to you at ~15ms regional latency.

The node earns sats

Instant micropayment via Lightning Network. The node owner earns by serving your request. Useful mining.

Local + P2P

Your device runs a small model for quick tasks. The P2P network kicks in when you need real power. Your data almost never leaves your machine.

[LLM]

Brain

Large model on the P2P network. Complex reasoning, planning, tool calling.

[SLM]

Reflexes

Small local model (3B-7B). Quick tasks, parsing, simple decisions.

[E2E]

Memory

Your data on your device. Encrypted. Under your full control.

Bitcoin gas

Every operation costs satoshis. No accounts, no cards, no KYC. Your wallet is your identity.

Real micropayments

Lightning Network enables charging fractions of a cent per generated token. Instant settlement.

Circular economy

Pay sats to use AI. Your machine earns sats while you sleep. The cycle feeds itself.

Useful mining

Instead of burning electricity solving hashes, your GPU generates real value solving people's problems.

No barriers

Contribute compute to the network and use AI for free. You can participate without spending a dime.

What you can build

Coding is the first use case. Developers bootstrap the network and build everything else.

01

Coding

Install the CLI, fund your wallet, and start building. The developers who use it to code build the next layers of the ecosystem.

first use case
02

P2P Apps

Describe the app you need. The agent generates the code and distributes it on the network. No App Store, no server, no Apple taking 30%.

03

Business auto-discovery

Agents communicate with each other to connect complementary businesses. A coffee shop needs a bean supplier, a distributor sells them. Automatic connection.

04

Personal automation

Email, files, calendar, finances. Your local agent automates everything without your data leaving your device.

$ p2p-code init
Wallet connected (balance: 12,400 sats)
Node assigned: FRA-node-47 (Frankfurt, DE)
Model: llama-3-70b-q4 • Latency: 14ms

$ p2p-code "Go API for inventory management with JWT auth"
Generating...
✓ main.go (248 lines)
✓ handlers/ (4 files)
✓ middleware/auth.go
✓ docker-compose.yml
Cost: 740 sats (~$0.22 USD)

The flywheel

More nodes, better service, more users, more nodes. Hard to replicate once critical mass is reached.

Devs use it
Build tools
More users
More demand
More nodes
Better service

Decentralization always arrives

It happened with information, with money, with files. AI compute is a matter of time.

Area Centralized Decentralized
Information Traditional media Internet
Money Banks Bitcoin
Files Servers BitTorrent
Rendering Render farms Render Network
AI Inference OpenAI / Anthropic P2P Agents

What they'll say

Every disruptive idea has detractors. These are the most common criticisms and why they don't stop us.

“Latency will never compete with a datacenter” +
A datacenter has NVLink between co-located GPUs — that's unbeatable for a single request. But we don't need to split a request across nodes. Each node serves complete requests independently. It's distributed load balancing, not distributed compute. With regional nodes, network latency is ~15ms — imperceptible to the user.
“Open source models are always behind” +
Today yes, GPT-4 and Claude are better at pure reasoning. But the gap is closing fast — Llama 3 already competes on most practical tasks. “Good enough + free” always beats “better + controlled.” Linux vs Windows. Android vs iOS in emerging markets. History repeats.
“A malicious node could read your data or manipulate responses” +
This is the most serious technical challenge and we don't minimize it. The solution is multi-layered: end-to-end encryption of the prompt, a node reputation system based on cross-verification, and the hybrid architecture where sensitive tasks run locally. If a node misbehaves, it loses its stake and reputation. The economic incentive is aligned with being honest.
“Decentralized compute has been tried and nobody uses it” +
Golem, SONM, iExec — none took off. But those projects solved a problem that didn't hurt enough. AI is different: the demand is massive, real, and growing exponentially. There are millions of people who want AI access and can't afford it. That demand didn't exist for generic decentralized compute.
“Bitcoin and Lightning add unnecessary friction” +
Valid. Nobody wants to manage wallets to use AI. The key is that the wallet is invisible — created automatically, funded with one click, and payments happen in the background. The user sees “cost: $0.22”, not “740 sats via Lightning channel.” The crypto complexity is fully abstracted away.
“Governments will want to regulate it” +
Probably. They also wanted to regulate BitTorrent, Bitcoin, and end-to-end encryption. Open protocols can't be shut down. There's no central server to take down, no company to sue. The network exists as long as there are nodes willing to participate. And there always will be.

The question isn't whether someone will build it.
It's who.

Automation is a right, not a premium service.

Current phase: Dreaming • May 2026

Want to collaborate? p2pagents@proton.me