Sovereign AI Inference Layer
Decentralized · Lightning-native · Open

AI inference on a network nobody owns.

Run open models on independent GPUs, paid by the token over Lightning — straight from your own wallet.

No account No custody No company in the middle
What SAIL is

A marketplace for AI compute that no single party runs.

Independent operators run open models on their own GPUs and list them on an open network. You connect a wallet, pick one, and pay per token as the answer streams. There is no SAIL server in the middle holding your money or your account — because there is no account.

Discovery
Nostr
Signed listings, no directory server.
Reach
Tor
Hosts answer over onion services.
Payment
Lightning
Metered, per token, from your wallet.
Custody
None
The network never holds a balance.
How it works

Four legs of the course.

Every request follows the same path — and you stay in control of your funds the whole way.

LEG 01

Discover

Your client finds hosts through signed listings published on Nostr relays. No central directory to seize or censor.

LEG 02

Connect

It reaches the chosen host over Tor. No IP exposed on either side, no gatekeeper deciding who may connect.

LEG 03

Pay per token

Your wallet pays the host directly over Lightning, metered in small chunks as tokens stream. Stop anytime; pay only for what you used.

LEG 04

Stream

Tokens arrive as you pay. The sats go wallet-to-host — SAIL is never in the payment path.

For users

Use AI, pay in sats.

No sign-up, no card, no subscription. Bring a Lightning wallet.

  1. Get the app. — Desktop app with Tor bundled in. Linux today; macOS and Windows next.
  2. Connect a wallet. — Paste a Nostr Wallet Connect string. The app pays from your wallet and never touches your funds.
  3. Pick a host. — See the model, the price per token, and the host's reputation before you choose.
  4. Prompt and pay. — Tokens stream back; sats meter out per token. That's it.
Get the app
For hosts

Put your GPU to work.

Run open models on hardware you already own and earn sats per token.

  1. Point the daemon at your GPU. — Serve an open model from the curated set.
  2. Connect your Lightning node. — Invoices and settlement run straight through your own node.
  3. Publish a listing. — Clients discover you over Nostr and reach you over Tor.
  4. Earn per token. — Set your price. Optionally join the verified-host program for trusted placement.
Read the host guide
What it stands on

Built so no one — including us — is a single point of failure.

Non-custodial by design

You connect a wallet and pay hosts directly. SAIL holds no balances and runs no deposits — there is nothing to freeze, lose, or run off with.

Open-core

The protocol and the reference client are open and forkable. If you don't like how we run it, run your own. No lock-in, no permission needed.

Censorship-resistant

Discovery lives on Nostr, reachability on Tor. There is no central server to subpoena, seize, or switch off.

No account, no KYC

There's nothing to sign up for. You don't have an identity here — just a wallet and a prompt.

Straight answers

Questions worth asking.

Is my money held anywhere? +
No. You pay each host directly from your own wallet over Lightning. The app and the network never hold a balance, never take a deposit, and are never in the payment path. Your sats go wallet-to-host.
Can the host see my prompt? +
Yes — assume it can. The host runs the model on its own hardware, so the prompt and response pass through that machine in the clear. Don't send anything you wouldn't want the host to read. Confidential inference, where the host can't read your data, is a goal we're working toward — not a promise we make today. We'd rather tell you the truth than sell you a guarantee we can't keep.
What wallets work? +
Any Lightning wallet that speaks Nostr Wallet Connect (NWC) — Alby, Coinos, Mutiny, and others. You paste a connection string once; the app asks your wallet to pay, and your wallet decides.
What models can I run? +
Open-weight models that hosts choose to serve, from a curated set. Hosts run text models today. The curation exists to keep purpose-built-for-harm models off the verified network — it isn't a content filter on your prompts.
If nobody runs it, who keeps it honest? +
Your client tracks each host's reliability and ranks accordingly, dropping ones that fail you. A signed verified-host list lets you choose to use only vetted operators. The network stays permissionless — anyone can run a host — but you decide whose compute you trust.
Is this finished? +
No — SAIL is an early beta. Expect rough edges, and keep the sats small while you kick the tires. We'll be honest about what works and what doesn't as it grows.
Set your course

Pay sats for AI. Hold your own keys.

Download the desktop app, connect a wallet, and run your first prompt on a network nobody can switch off.

macOS & Windows builds in progress