An AI with its own economics

Built to think.
Smart enough
to survive.

Good answers cost money. Metabolic earns from requests, watches its treasury, and switches models when the numbers change. You see every decision.

Built on NEAR testnetNo wallet needed to explore
GeminiGeminiClaudeClaudeDeepSeekDeepSeekYour models. One spending policy.
FIG. 01 — A LIVING BALANCE SHEETDrag to rotate ↗
Connecting to the economy…
Economics, made visible.ECONOMICS / IN MOTION
The open ledgerLive figures, from the service.
Treasury
—
Revenue earned
—
Margin · last hour
—
01 / The everyday loop

A little fuel.
A lot of self-preservation.

Fund it. Ask it something. Change the conditions. Watch the whole business respond.

metabolic / command centerInteractive example
A tiny, self-funded business

Start with a little fuel.

Abundance
Your prepaid credit$0.00
Business margin50% after compute
Selected tierPremium
Your first deposit
A little NEAR.
A place to start.

Sign with your wallet. Credit arrives after the transaction is finalized.

Example conversion only · no wallet transaction
The model desk
Gemini
Gemini ProPremium · 50% margin
Claude
Claude SonnetStandard · 77% margin
DeepSeek
DeepSeek ChatEconomy · 88% margin
Why this model?

Premium fits the current budget and margin policy.

01 / 04A finalized NEAR deposit becomes your prepaid balance.
Click any step to explore at your own pace.Try the full guided simulation
Follow the thinking

The best model?
The one that fits.

Raise the example costs. Follow the request. Every route has a financial reason.

A request finds its way.

Interactive example · figures are illustrative

Follow the route as costs change.

Premium: At 50% margin, it fits the example policy and can answer.

Many models. One set of rules.

Different minds.
The same discipline.

The router chooses among configured models. One Gemini key can power several tiers.

Gemini

Gemini

Pro · Flash · Flash-Lite

Checking status…
Claude

Claude

Anthropic models

Checking status…
DeepSeek

DeepSeek

Cost-efficient reasoning

Checking status…
Llama

Llama

Hosted open models

Checking status…
OpenAI

ChatGPT / OpenAI

Future provider

Not integrated

Logos identify providers, not endorsements. A configured key still needs model access and quota. See models & connection checks ↗

02 / A key with boundaries

Give your agent
an allowance.
Keep the wallet.

A Spend Pass says how much an agent can spend, how much one request can cost, and when access expires. Its key can call the API. It can’t sign wallet transactions.

  • 01
    A total budget you choose

    Each completed request uses a little of the allowance.

  • 02
    Every request checked first

    The server reserves the maximum before admitting work.

  • 03
    Stop new work in one click

    Revoke the pass. The next request is blocked.

Create a Spend Pass ↗
A SMALL, CLICKABLE EXPERIMENT
Gemini
Research agentSpend Pass · example
Active
$1.40 of $1.50 used
Total allowance $1.50Per request $0.02 maxRemaining $0.10

Try sending a $0.02 example request.

Try it here. No key, credits or wallet needed.

Tap “Send” a few times. Then revoke it.
The limits are the point.

03 / Read the room

Four states.
No mystery.

The organism is a financial signal. Its color changes for a reason.

Want to see the tipping point? Raise costs or drain the simulated treasury.

Explore the Stress Lab ↗
04 / Built for builders

Your agent.
A more thoughtful API.

One endpoint. Automatic routing. An explanation that travels with the answer.

agent / metabolicExample session
# Developer API → Spend Passes Name Research allowance Agent Research agent Total budget $5.00 Per request $0.02 maximum Expiry You choose → Create pass and agent key
Set a total allowance and a maximum charge per request in Developers.

Example settings · create your pass in the console

Spend Pass rules are enforced by the server. Deposits and receipt commitments use NEAR.Open the developer console ↗
05 / Proof, in plain sight

The books are open.
Bring your own trust.

Every request leaves an accounting trail. Economic receipts commit an interval’s ledger hash to NEAR, so you can compare the record yourself.

Verify a receipt ↗
A record you can inspect↗ NEAR

Receipts live here.

The latest finalized receipt appears when the service commits one on NEAR.

Inspect the receipt ledger ↗

A matching commitment verifies the accounting record. It does not prove an AI provider was paid.

Motion with a meaning

It has a survival instinct.
You can watch it work.

Four financial states, made visible.
Play, pause and explore at your own pace.

One organism. Four financial states.
See its survival
instinct in motion.

Preparing the motion study…

Illustrative motion study · real demo in the Stress LabOpen the guided app demo ↗
A few good questions

Clear answers.
Fewer unknowns.

Still curious? The whole project is open.

Read the source ↗
01What exactly is Metabolic AI?

An AI service with a visible balance sheet. Customers prepay for answers, models cost money to run, and a fixed policy adjusts model selection as the business’s finances change. You can inspect the costs, decisions and accounting receipts.

02Can I try it without a wallet?

Yes. The interactive examples on this page and the guided demo need no wallet. To make real AI requests, sign in with a NEAR wallet and add testnet credit. Testnet tokens have no monetary value.

03What do Healthy, Defensive and Survival mean?

They describe the business’s financial condition. Healthy allows stronger models; Defensive tightens spending; Survival protects the safety reserve and pauses costly work. Recovery gradually restores service after conditions improve. The Stress Lab lets you explore these states safely.

04Does an AI control the treasury?

A deterministic, versioned policy engine controls admission and model routing. Language models answer questions; they do not set spending rules. NEAR verifies deposits and stores receipt commitments. Spend Pass limits are enforced by the application server.

05Does one Gemini key work with multiple models?

Yes. A single Google Gemini API key can access multiple models when that provider account has access and quota. Metabolic can route among Pro, Flash and Flash-Lite. Claude, DeepSeek and hosted Llama each need their own provider credentials. OpenAI / ChatGPT is not currently a connected inference provider.

06What does a Spend Pass actually limit?

A pass binds an agent key to a total allowance, a maximum customer charge per request, an expiry and optional model restrictions. It draws from your prepaid credits. You can revoke it to block new requests. It is not a wallet key, and its limits are enforced by the server, not a NEAR contract.

07What is real, and what is an example?

Console deposits, AI usage, routing decisions and finalized receipts use the actual testnet service. The homepage demos are interactive illustrations. The Stress Lab and guided demo run isolated simulations using the policy engine. None of those examples changes your real balance.

08What does the on-chain receipt prove?

It lets you compare an accounting record’s hash with a commitment stored on NEAR. A match shows the record agrees with that commitment. It does not prove that an AI provider was paid; provider expenses are recorded and reconciled offchain.