Robinhood Chain · 4663
A flat balance cannot be delegated. Give an agent one and it can spend all of it, on anything, forever. Metronium gives it a grant instead: an amount with rules attached that travel with the token.
The problem
One balance, one key, no boundaries. That is fine when a person buys inference for themselves. It stops being fine the moment the spender is an agent running on someone else's schedule.
Hand an agent your key and it can reach every model you can, at any rate, until the money is gone. The only ceiling is your entire account.
Send 50 METRON that may only buy one model, expires in seven days, burns no faster than $5 an hour, and comes back to you if the agent stops working.
How it works
Buy it on the open market. Fixed supply, no mint function, no owner — the contract can do nothing except be an ERC-20.
Send METRON to the treasury address. Your API balance is credited in dollars at the time-weighted rate, and that dollar figure never moves again.
Lock an amount for a recipient with a policy: allowed models, expiry, hourly dollar cap, and whether you can take it back.
Two environment variables. The gateway checks every request and answers 402 the moment a limit is reached.
# Point any OpenAI-compatible client at Metronium. OPENAI_BASE_URL=https://api.metronium.xyz/v1 OPENAI_API_KEY=sk_mtr_… # Model names, streaming, tool calls and vision behave exactly as before.
Delegation
A coordinator gives each worker its own grant. A DAO pays a contributor in inference. A platform funds a user's agent without ever sharing the provider key. Same call every time.
// METRON uses 6 decimals: 1 token = 1_000_000 units. metron.approve(address(grants), 50_000000); grants.grant(worker, 50_000000, Policy({ modelsRoot: allowlistRoot, // only these models expiresAt: uint64(block.timestamp + 7 days), ratePerHour: 5_000000, // at most $5 per hour revocable: true // unspent value can come back })); // Worker stopped early? Take back whatever it never activated. grants.revoke(grantId);
Auditable, not trustless. Rate caps are enforced by the gateway, because a chain cannot know how much was burned during the last hour. What lives on chain is the commitment — anyone can check afterwards that we honoured it. We will never market this as trustless.
Commitments
The response carries the provider's own model id. Ask for Opus, read Opus back off the payload.
Model, token counts, cost, timing. Nothing else is written — not to our disks, and never to the chain.
Break a policy and the next request is a 402. Not an email the following morning.
X-Metron-Cost and X-Metron-Remaining on every reply. An agent can budget itself.
$METRON
METRON pays for inference across 443 models. Fixed supply, no mint function, no owner, no yield. Demand comes from people using the product, not from an emission schedule — which is the only kind of demand that survives a flat market.
Launching 18 September 2026, 16:00 UTC on Pons. The contract address goes up on Contracts the moment it is live. Verify there and nowhere else.
| Property | Value |
|---|---|
| Ticker | METRON |
| Supply | 1,000,000,000 — minted once, never again |
| Decimals | 6 |
| Price | Floating. Inference stays priced in dollars. |
| Owner / mint / pause | None of them exist on the contract |
| Chain | Robinhood Chain · 4663 |
| Launch | 18 Sep 2026, 16:00 UTC · Pons |