LANEX
Docs
Rev 1 · 14 Nov 2026. This describes the early version. Things will change before the network is live.
01Overview
LANEX is a market for AI inference. People with GPUs host models and post a price per 1M tokens. People who need inference send calls through a router on their own machine, and the router picks a seller for each model based on a rule you set.
Payment comes out of an escrow balance on Robinhood Chain. You choose how much you're willing to spend per day, and when you hit that number the router stops sending paid calls.
02Routing a call
Rules
| Rule | What it does |
|---|---|
cheapest | Lowest price among sellers that are online and below their rate limit. This is the default. |
fastest | Lowest recent latency. Measured over the last few minutes of calls. |
pinned | One named seller. If they're offline the call fails with SELLER_OFFLINE and doesn't fall back. |
Daily cap
The cap is in USD and resets at 00:00 UTC. A call that would go over it is refused before it's sent, so you never end the day above your number. Refused calls don't pay a fee.
Seller limits
Each seller sets a rate limit (requests per minute) and a daily token limit. When a seller is at either one, the router skips them as if they were offline.
Errors
| Code | Meaning |
|---|---|
NO_WALLET | Not signed in, or the signature didn't check out. |
NO_DEPOSIT | Escrow balance is zero. |
CAP_REACHED | You've spent your daily cap. |
SELLER_OFFLINE | The pinned seller (or the chat model) isn't reachable. |
BUSY | Every seller for that model is full, or you hit a rate limit on our API. |
BAD_ARG | Something in the request was missing or out of range. |
03Market board
The board lists every offer for a model: seller name, price per 1M tokens, rate limit and current load. The chart groups trades into 20 second buckets and shows the average price paid and the volume in tokens. The tape on the right is the last few calls.
Offers are sorted by price. Offline sellers stay on the list, greyed out, so you can see who usually shows up.
04Fees
LANEX takes 0.825% of each call that settles. It's the same for every model and every seller.
| Calls | Fee |
|---|---|
| One $1 call | $0.00825 |
| $100 of calls | $0.825 |
| Refused call | $0 |
Gas on Robinhood Chain is paid in ETH and is separate from the fee.
05Wallet and network
| Field | Value |
|---|---|
| Network | Robinhood Chain |
| Chain ID | 4663 (0x1237) |
| Gas token | ETH |
| RPC | https://rpc.mainnet.chain.robinhood.com |
| Explorer | robinhoodchain.blockscout.com |
Sign-in is a plain text message you sign in your wallet. It proves you own the address. It doesn't move funds, grant approvals or let us do anything with your wallet.
06Bridges
You need ETH on Robinhood Chain for gas. These routes support it. We don't run any of them, so check fees and limits on their side.
- Arbitrum canonical bridge
- Across
- Relay
- Stargate
- Transporter (Chainlink CCIP)
- LI.FI and 0x aggregators
07Selling capacity
You run the model, we route calls to it. A listing has five fields:
| Field | Allowed |
|---|---|
| Display name | Up to 40 characters |
| Model | One of the listed models |
| Price | 0.001 to 100 USD per 1M tokens |
| Rate limit | 1 to 10,000 requests per minute |
| Daily token limit | 1,000 to 10,000,000,000 |
You get paid per settled call, minus the fee. If nobody routes to you, you earn nothing, so price against the board. The model names on the site today (chat-70b, chat-8b, code-34b, vision-11b, embed-1b) are placeholders.
08Developers
Local router
Still being built. It will expose /v1/chat/completions and /v1/embeddings on localhost with the same shapes as the OpenAI API, plus an optional policy field.
Web API
Same origin as the site. JSON in and out.
| Route | What it does |
|---|---|
GET /api/health | Status check |
GET /api/auth/nonce?address=0x… | Returns the sign-in message for that address |
POST /api/auth/verify | Send {message, signature}, get a session cookie |
GET /api/me | The address you're signed in as, or null |
POST /api/auth/logout | Clears the session |
POST /api/receipt/sign | Signs a call receipt |
POST /api/receipt/verify | Checks a receipt signature |
POST /api/chat | Chat, signed-in only |
POST /api/whitelist | Join or leave the list |
curl https://YOUR-HOST/api/auth/nonce?address=0xYourAddress
Limits: 120 requests per minute per IP across the API, 20 per minute on auth, receipt, chat and list routes. Bodies over 8 KB are rejected.
09Token and Pons
There is no LANEX token yet. No date and no contract address have been set.
The plan is to launch on Pons, a launchpad on Robinhood Chain. A Pons token starts on a bonding curve where you buy and sell with ETH. When the curve holds 4.2 ETH it graduates and the liquidity moves to a Uniswap V4 pool.
10Your data
- Signing in sets a cookie with your address. It lasts 24 hours or until you sign out.
- The early access list stores your email and the time you joined. "Remove my email" deletes it.
- Chat messages go to the model provider to get a reply. We don't store them.
- No analytics cookies and no tracking pixels.
11Risks
- This is early software and nobody has audited it.
- Market data on the site is simulated until the network launches.
- Sellers can go offline, return bad output or stop responding. Pinned routes fail when that happens.
- A token, if launched, can go to zero. Bonding curve prices move fast in both directions.
- Nothing here is financial advice.
12FAQ
Do I need a wallet to look around?
No. The playground and the board work without one. Chat needs a signed-in wallet so we can rate limit it.
Can I use my existing OpenAI key setup?
Yes. Change the base URL to the local router and keep the rest of your code.
What happens if a seller lies about the model?
Every call gets a receipt you can check, and sellers with bad output lose traffic fast. Proper verification of model output is on the list for after launch.
When does the token launch?
Not decided. Watch the site and the Telegram channel.