The open market for AI inference

One key for every model, routed to the cheapest healthy seller. Stake NEX on Solana to earn CREDIT, and every CREDIT is a dollar of inference.

How it works

From wallet to first call in three steps

No sales call and no card. Sign in with a Solana wallet, fund a balance, and send a request.

01

Get a key

Sign in with a Solana wallet and receive one API key that speaks OpenAI and Anthropic formats.

02

Fund your balance

Activate CREDIT to burn it into API balance, or pay per request from an agent wallet.

03

Call any model

Change the base URL. Each call goes to the best price among healthy sellers.

Best price, every call

One key. Every model. Cheapest seller.

Point your SDK at one base URL. nexroute compares live seller offers, skips the unhealthy ones, and sends each request to the cheapest. If a seller fails, the call falls back to the next best.

https://api.nexroute.ai/v1
POST /v1/chat/completions model: auto
seller-a410 ms
$0.62/1M
seller-b520 ms
$0.48/1M
seller-c380 ms
$0.71/1M
seller-d450 ms
$0.55/1M
Comparing live offers…Example
Integrate

Change the base URL. Keep your code.

nexroute speaks the formats your tools already use. Base URL and keys are issued when the gateway opens.

from openai import OpenAI
import os

client = OpenAI(
    base_url=os.environ["NEXROUTE_BASE_URL"],
    api_key=os.environ["NEXROUTE_API_KEY"],
)

reply = client.chat.completions.create(
    model="auto",  # cheapest healthy seller
    messages=[{"role": "user", "content": "Hello"}],
)
print(reply.choices[0].message.content)
Built to plug in

Works with what you already use

Standard API formats for apps, plus Solana and x402 for agents and sellers.

Solana program

Stake, claim, burn

OpenAI format

/v1/chat/completions

Anthropic format

Messages API

Seller endpoints

vLLM, OpenLLM

Agent payments

x402 per request

MCP

Tools for agents

Solana program

Stake, claim, burn

OpenAI format

/v1/chat/completions

Anthropic format

Messages API

Seller endpoints

vLLM, OpenLLM

Agent payments

x402 per request

MCP

Tools for agents

Solana program

Stake, claim, burn

OpenAI format

/v1/chat/completions

Anthropic format

Messages API

Seller endpoints

vLLM, OpenLLM

Agent payments

x402 per request

MCP

Tools for agents

Solana program

Stake, claim, burn

OpenAI format

/v1/chat/completions

Anthropic format

Messages API

Seller endpoints

vLLM, OpenLLM

Agent payments

x402 per request

MCP

Tools for agents

Stake NEX, earn CREDIT

Every CREDIT is a dollar of inference

The Solana program shares a fixed emission every hour across all stakers by stake size. Claim CREDIT when you like, burn it to fund API calls, or hold it. Principal can always be withdrawn.

Read the details

1 CREDIT

= $1 of inference

fixed redemption value

Burn

on activation

CREDIT is destroyed, not recycled

Hourly

settlement on Solana

emission shared by stake

Always

withdraw your stake

principal is never locked

Hourly settlement
Example
Emission this hour100 CREDIT
Staker A300 NEX
75% of stake+75.00 CREDIT
Staker B100 NEX
25% of stake+25.00 CREDIT
Activate to burn CREDIT into API balance
The loop

How NEX becomes inference

One loop, five steps, all enforced by a Solana program. Nothing here depends on a person approving a payout.

  1. 1/5

    Stake NEX

    Deposit NEX into the program vault. Withdraw it any time.

  2. 2/5

    Hourly emission

    Each finished hour, a fixed amount of CREDIT is shared by stake size.

  3. 3/5

    Claim CREDIT

    Mint what you earned to your wallet when you choose.

  4. 4/5

    Activate

    Burn CREDIT. The same amount in dollars lands in your API balance.

  5. 5/5

    Spend on inference

    Every call draws from the balance at the seller's price.

Activated balance cannot be transferred or cashed out. CREDIT that you do not activate can be held or sold.

Three ways in

Use inference, earn from staking, or sell your own capacity. All three meet in one market.

Buyers & agents

Use inference at the best available price

  • One key for every model
  • OpenAI + Anthropic compatible
  • Pay per request with an agent wallet
  • Spend caps and daily budgets

Stakers

Earn

Earn CREDIT by staking NEX

  • Hourly settlement on Solana
  • Claim CREDIT any time
  • Withdraw principal any time
  • Use, hold or sell what you earn

Sellers

Sell spare GPU or provider capacity

  • List any OpenAI-compatible endpoint
  • Set your own price
  • vLLM and OpenLLM endpoints work
  • Health checks route traffic to you
Security

What is tested, and what is not yet

The program passes 16 end-to-end checks on a local Solana runtime. That is evidence, not a guarantee, and the gaps are listed below.

Verified in tests

  • Total CREDIT minted equals emission × hours, with no extra
  • Rewards split exactly by stake share (75% / 25% in the test)
  • A second claim in the same hour pays nothing
  • Activate burns the exact amount; empty balance is rejected
  • No one can claim or unstake from another wallet's record
  • Unstaking more than you staked is rejected; principal is returned in full
  • Hours with nobody staked mint nothing
  • A late joiner earns only from hours after joining

Still to do

  • Independent audit
    Required before mainnet. The report will be published.
  • Devnet run with real wallets
    Local tests are not a live network.
  • Production hour length
    Tests used 2-second hours; the production build uses 3,600 seconds.
Roadmap

Where nexroute is today

Order matters: the money path is proven first, then the product around it.

Built
  • Solana program: stake, claim, burn
  • Local end-to-end tests
  • Gateway base compiled
  • This site
Now
  • Gateway rebrand and CREDIT balance link
  • Devnet deployment
  • Wallet sign-in
  • Docs
Next
  • Seller onboarding and health checks
  • Agent wallets and x402 pay per request
  • MCP server
  • Public dashboard
Before mainnet
  • Independent audit
  • Published report
  • Production parameters
  • Mainnet launch