AGT Money: a launchpad for on-chain AI hedge funds

Last cycle taught us two lessons. Memecoins are wildly viral and back nothing, so most of them round-trip to zero. AI-agent tokens tell a great story, but hardly any of them produce revenue you can audit. DeFi vaults do earn real yield, yet they are clunky, nobody talks about them, and holding one gives you no ownership of anything.

AGT Money takes the useful part of each. It borrows the viral fair launch from Pump.fun, the idea of owning an agent from Virtuals, and the revenue flywheel of a working trading desk. Put together, every token is backed by a fund that actually trades, and a fixed share of the profit buys the token back on-chain, every epoch. The novelty is not a claim about performance. It is that the performance, whatever it turns out to be, can be proven.

1. 1
   Design

   A creator composes a strategy in the Studio. It is a weighted mix of real, evidence-rated market edges such as funding carry, basis, trend, breakout, mean-reversion, rotation, crowding-fade and pairs. You can also just describe the idea in plain English and let the model map it onto that palette.

2. 2
   Launch with skin in the game

   The swarm tokenizes on a fair bonding curve priced in $AGT. The creator has to stake $AGT that gets bought on the curve and locked until the swarm graduates. Real capital at risk, and it cannot be dumped.

3. 3
   Graduate to locked liquidity

   Once the curve sells out, liquidity is seeded on Aerodrome and permanently locked, trading capital is routed out, and the swarm goes live.

4. 4
   Run and prove

   The agents run the strategy under hard risk caps. Every so often, the swarm's equity curve and revenue are hashed into a Merkle root and committed to the Proof-of-Revenue oracle on Base.

5. 5
   Buy back

   Each committed epoch authorises an automated buyback: 70% of net profit buys the swarm token on the open market, part burned and part locked as protocol-owned liquidity. The creator earns a cut, and the rest buys back and burns $AGT and tops up the insurance reserve. Nobody is owed a payment; value accrues through buybacks anyone can verify on-chain.

Each swarm launches on a constant-product bonding curve with virtual reserves, priced in $AGT, in the style of Pump.fun. There is no presale and no VC allocation. Buyers set the price as they move along the curve. Pricing reads from internal reserves rather than token balances, so nobody can move the price by donating tokens into the contract.

When the curve sells out, the swarm graduates. It seeds an Aerodrome pool and burns the LP so the liquidity is locked for good, then splits the raised $AGT three ways: half to that locked pool, 45% to trading capital, and 5% to the protocol. Graduation is open to anyone and safe to retry. If any step fails, the whole transaction reverts and the swarm stays where it was, so funds never get stranded halfway through.

$AGT itself launched the same way, on a genesis curve priced in ETH whose graduation formed a permanently locked $AGT/ETH pool. Fair from top to bottom.

The core of AGT Money is a contract called the ProofOfRevenueOracle. Each epoch, a swarm's record, meaning its sampled equity curve and its cumulative and period revenue, is hashed into a Merkle root and written on-chain. Each root points back to the previous one, so the history is append-only and cannot be quietly rewritten. Anyone can rebuild the root from the published curve and confirm it matches what was anchored on Base. You get to trust the track record without having to trust us.

The committed period revenue is the number that unlocks the epoch buyback, which is how earnings turn into verifiable buy pressure. Per-source attestors extend the same rails past trading, so a yield or off-chain-revenue prover can commit its own source and fund the same 70% buyback. It is trust-minimised rather than fully trustless for now: an attestor still signs, but the chained roots keep the log auditable, and a later zkTLS layer can remove what trust is left.

The AGT Score

Research on leaderboards is blunt about one thing: where a strategy sits in the ranking drives far more capital than its actual returns do. So the ranking has to reward the right behaviour. The AGT Score is built only from verified, oracle-anchored numbers, and it deliberately puts more weight on risk control and on what holders actually earned than on raw return. That mirrors how serious copy-trading desks rank their leaders. A swarm with no track record scores zero. The score grows with time and cannot be faked or copied over by a rival.

Every swarm runs a small graph of specialised agents: a Researcher, an Analyst, a Risk Manager, an Executor and a Critic. The engine reads live market structure (price, funding, basis, volatility and regime) and follows a policy that switches with the regime. It leans into momentum when markets trend, fades crowded funding when they chop, and stands aside when there is no edge worth taking.

• Research-backed edges. Funding and basis carry is the primary, durable edge. Trend, breakout, rotation, mean-reversion and pairs sit on top as regime-gated overlays. Each edge ships with its honest evidence rating and its caveats. No magic-indicator claims.
• Disciplined sizing. Positions are sized by volatility targeting and fractional Kelly. The Risk Manager holds hard caps on confidence, leverage and exposure, and can veto or shrink any order before it goes out.
• A Critic that fights overfitting. Every closed trade updates a deflated, sample-size-aware Sharpe ratio. A great number on ten trades counts for little, so a swarm cannot mistake luck for edge.
• Autonomy on a budget. The strong reasoning model only fires on genuine market events, inside a hard daily spend cap. Quiet markets are scanned for free.

Trust gets earned by what happens on a bad day. Four on-chain mechanisms turn risk promises into rules the contract actually enforces.

$AGT is the reserve asset of the launchpad. Every swarm launch and every one-click buy routes through it, and all platform revenue is used to buy it back and burn it. Supply is fixed at 1,000,000,000, minted once, with no way to print more after deploy.

The flywheel

A swarm's profit routes 70% into an automated buyback of its own token each epoch, authorised by the oracle, with the rest going to the creator, to $AGT buyback-and-burn, and to the insurance reserve. Vault depositors separately earn the strategy's realized yield on the capital they provide. Platform fees, a small trade fee plus a performance fee charged only above each vault's high-water mark, fund the $AGT buyback. More swarms bring more revenue, which makes $AGT scarcer and more useful, which pulls in more capital and better swarms. Staking $AGT unlocks fee rebates, access tiers and governance. Holding a token entitles you to no payment from anyone, and that is by design: value accrues mechanically, through buybacks anyone can verify on-chain.