← BACK TO $SIZE

DOCUMENTATION

Leveraged-PUMP-Infrastructure · Technical Reference · Updated April 2026


$SIZE is autonomous infrastructure built entirely on pump.fun's official Agent Skills SDK. It turns idle creator fees into productive, leveraged on-chain capital. Every cycle, the agent claims accumulated creator fees using the coin-fees skill, consults an AI brain for an entry decision, executes swaps via the swap skill, and opens a real spot-leveraged PUMP position on spot margin. Every action is logged. Every transaction is on-chain. Every reasoning is published in plain English.

This project is only possible because pump.fun released their Agent Skills — giving programmatic access to fee collection, sharing configs, token swaps, and payment verification that previously required manual interaction. $SIZE is a demonstration of what becomes possible when a platform opens its infrastructure to autonomous agents.

The flagship vault routes SIZE's own creator fees into leveraged PUMP longs because PUMP is the meme that started everything. But the engine is not hardcoded to PUMP — the target token, host token, leverage, and split are all configuration. SIZE is the pattern, sold as infrastructure, to any meme that wants it.

StatusLive · 2 active positions
Enginespot margin spot leverage (real PUMP)
BrainOpenAI GPT-4 class reasoning
CycleEvery 15 minutes
Min Claim0.05 SOL
Default Split70% leverage / 30% buyback
Default Leverage5× (brain-adjusted, max 7.5×)
Agent Wallet2i1i4UJBWKu9...nR6qxw

Pump.fun released an open-source collection of Agent Skillsthat teach AI agents how to interact with their on-chain programs. $SIZE uses all four skills in its pipeline. Before this SDK existed, building this pipeline meant reverse-engineering pump.fun's programs, manually resolving PDAs, handling bonding curve vs. AMM state transitions, and constructing raw Solana transactions from scratch.

coin-fees — Creator Fee Collection & Distribution

The core of the pipeline. Every 15 minutes, $SIZE calls the coin-fees skill to claim accumulated creator fees from connected token vaults. The skill automatically detects fee destinations — whether fees go directly to the creator, to a sharing config with multiple shareholders, or as cashback to traders. It resolves vault balances across both the Pump program and Pump AMM sides, and builds the correct claim transaction (collect for direct creator, distribute for sharing configs).

Automatic fee destination detection (creator / sharing config / cashback)
Vault balance calculation across Pump + AMM creator vaults
Permissionless — anyone can trigger fee collection
Sharing config support for up to 10 shareholders with basis-point splits
api-call
POST https://fun-block.pump.fun/agents/collect-fees
{
  "mint": "<TOKEN_MINT>",
  "user": "<AGENT_WALLET>",
  "encoding": "base64"
}

swap — Token Swaps with Auto-Detection

Used in two places: (1) the 30% buyback leg swaps SOL → $SIZE for burning, and (2) position entry/exit on $PUMP when needed. The swap skill automatically detects whether a token is still on the bonding curve or has graduated to the AMM pool, and builds the correct transaction type. Slippage protection and optional Jito front-runner protection are built in.

Auto-detects bonding curve vs. graduated AMM state
Configurable slippage protection (default 2%)
Optional Jito front-runner protection via tip mechanism
api-call
POST https://fun-block.pump.fun/agents/swap
{
  "inputMint": "So11111111111111111111111111111111111111112",
  "outputMint": "<TOKEN_MINT>",
  "amount": "1000000",
  "user": "<AGENT_WALLET>",
  "slippagePct": 2,
  "encoding": "base64"
}

create-coin — Token Creation with Tokenized Agent Support

$SIZE itself was launched using the create-coin skill with the tokenized agent flag enabled. This means the token natively supports pump.fun's built-in buyback mechanism — a portion of every trade automatically flows back into the protocol. The skill handles mint keypair generation, initial buy, metadata upload, and partial signing in a single transaction.

Tokenized agent flag enables automatic buyback on every trade
Configurable buyback percentage via buybackBps (basis points)
Supports mayhem mode, cashback, and front-runner protection options
api-call
POST https://fun-block.pump.fun/agents/create-coin
{
  "user": "<CREATOR_WALLET>",
  "name": "SIZE",
  "symbol": "SIZE",
  "uri": "<METADATA_URI>",
  "solLamports": "1000000000",
  "tokenizedAgent": true,
  "buybackBps": 5000,
  "encoding": "base64"
}

tokenized-agents — On-Chain Payment Verification

For premium vault features (custom leverage targets, priority execution), $SIZE uses the agent-payments-sdk to build payment invoices and verify them on-chain. This enables a trustless paywall — no backend database of who paid, no API keys. The blockchain is the receipt. Supports both USDC and wrapped SOL as payment currencies.

On-chain invoice creation and payment verification
Supports USDC (6 decimals) and wrapped SOL (9 decimals)
Server-side verification via validateInvoicePayment() — never trust client alone
sdk-usage
import { buildAcceptPaymentInstructions, validateInvoicePayment }
  from '@pump-fun/agent-payments-sdk';

// Build payment instruction
const ix = buildAcceptPaymentInstructions({
  agentTokenMint, currencyMint, amount, invoiceId, startTime, endTime
});

// Verify on-chain (server-side)
const valid = await validateInvoicePayment(
  connection, agentTokenMint, currencyMint, invoiceId
);
Source: github.com/pump-fun/pump-fun-skills — Open source, MIT licensed. Each skill follows the Agent Skills format and can be loaded by any compatible AI agent.

Creator fees are the most underused primitive on Solana. Billions of dollars of volume flow through pump.fun. The creator fee system extracts a slice of every trade. That slice — across the entire ecosystem — represents an enormous stream of passive SOL that currently gets claimed and dumped. Friction on the engine of the meme economy.

SIZE is a proof that this does not have to be the default. Creator fees can be productive capital. They can compound into positions. They can fund buybacks. They can support the underlying memes that birthed them. And with the leveraged-pump-infrastructure model, any token can plug into the same infrastructure without rebuilding the wheel.

The pattern

A creator wallet accumulates fees in SOL.
An autonomous agent claims, splits, and routes those fees.
A configurable percentage goes into leveraged long exposure on a target token.
A configurable percentage goes into buyback and burn of the host token.
An AI brain decides entry sizing, leverage, and exits in real time based on live market data.
Every action is on-chain, logged, and verifiable.

The target token is a config variable. The host token is a config variable. The split is a config variable. The venue is a config variable. Swap PUMP for any token listed on a perpetual DEX or spot leverage protocol and the engine just works.


The pipeline is a deterministic loop running every 15 minutes. Each cycle has the same structure: check existing positions for exits, claim new fees, consult the brain, open a new position if the brain approves. No human intervention required after initial setup.

Stage 0: Exit monitoring

Brain reads each existing position and decides HOLD, PARTIAL_CLOSE, or FULL_CLOSE
Decision is based on ROI, momentum, distance to liquidation, and interest cost
Closed proceeds get unwrapped (WSOL → native SOL) and feed the next cycle

Stage 1: Claim

Agent monitors creator fee wallet via Solana RPC
Claims accumulated fees when balance exceeds 0.05 SOL threshold
All claims logged to Redis with transaction signatures

Stage 2: Split

70% allocated to leveraged PUMP position (configurable per vault)
30% allocated to buyback reserve for $SIZE burn
Split ratios are deterministic and visible in every cycle log

Stage 3: Brain consult

Brain receives live market context (price, 1h/24h change, existing positions)
Returns structured decision: OPEN at X× leverage, or SKIP, with reasoning
Reasoning published in plain English on the dashboard

Stage 4: Open

SOL collateral deposited directly to spot margin
spot margin borrows SOL against collateral and buys real PUMP tokens on-chain
Position address registered to Redis for dashboard tracking
Open transaction signature linked from /api/position
cycle.pseudo
// Every 15 minutes, deterministic
exit_monitor.check_all_positions()      // Stage 0
fees = claim_creator_fees()              // Stage 1
if fees < 0.05 SOL → log SKIP, return

long_amount    = fees × 0.70             // Stage 2
buyback_amount = fees × 0.30

market   = fetch_market_context()        // Stage 3
decision = brain.decide_entry(
  long_amount, market, existing_positions
)
if decision.action == 'SKIP' → return

  collateral: long_amount,
  leverage:   decision.leverage,
  asset:      PUMP
})

log(position, redis)
register_open_tx(position, signature)

Most yield protocols are dumb loops. Claim, swap, deposit, repeat. That is fine when the yield source is stable and the exposure is delta neutral. That is not fine when you are taking leveraged directional positions on a memecoin.

The brain is the difference between an autopilot and a thermostat. Every cycle it reads live market context, returns a structured decision with explicit reasoning, and logs both to Redis for the dashboard to display in real time. Anyone can read the brain's reasoning for every decision it has ever made.

Inputs

Current PUMP price (Pyth oracle / DexScreener)
1h price change, 24h price change, recent volatility
Existing position state: ROI, distance to liquidation, daily interest cost, collateral, effective leverage
Available SOL for new entry sizing
Configured leverage bounds (default: 5×, max: 7.5×)

Entry decision schema

brain.entry.json
{
  "action": "OPEN" | "SKIP",
  "leverage": 4.5,
  "reasoning": "PUMP showing positive 24h momentum with manageable
   volatility. Sizing at 4.5× to leave headroom on liquidation
   distance given 1h pullback. No existing position degrades to
   risk-off."
}

Exit decision schema

brain.exit.json
{
  "action": "HOLD" | "PARTIAL_CLOSE" | "FULL_CLOSE",
  "reasoning": "ROI is +4.76%, within HOLD range. Position is not
   under immediate threat of liquidation, interest cost is manageable,
   no strong momentum indicators suggest taking profit or cutting loss.",
  "urgency": "low" | "medium" | "high",
  "roi": 4.76,
  "distanceToLiq": 13.157
}

Live brain reasoning (pulled from /api/logs)

"The current ROI is +4.76%, which is between -15% and +15%. The position is not under immediate threat of liquidation, and the interest cost is manageable. No strong momentum indicators suggest taking profit or cutting loss."
Brain · Position 7jsFgFQb · HOLD
"The position has a positive ROI of 4.71%, which is within the HOLD range. The current market conditions show a slight upward trend, and the distance to liquidation is safe at 13.1%."
Brain · Position BeN8QMjM · HOLD

Why a brain instead of fixed thresholds

Holding through a 6% drawdown when support is nearby is a feature, not a bug
Cutting at 18% when momentum collapses is a feature, not a bug
A dumb loop for PUMP breaks the first time you point it at a different token with different volatility
The same brain that runs the PUMP vault can run a WIF vault, POPCAT vault, or token launching tomorrow with no code changes

spot margin provides spot leverage — borrowing SOL against your collateral and using it to buy real tokens on-chain. Unlike perpetual futures (synthetic exposure), this creates actual buy pressure on the underlying asset.

Protocolspot margin
TypeSpot leverage (real tokens)
PUMP Pools9 active
Max Leverage7.47×
Borrowing APR~99%
Liquidation ModeSELL (auto-unwind)
CollateralSOL (direct deposit)

Why spot margin over Drift

Real buy pressure — spot margin buys actual PUMP tokens vs. Drift's synthetic perpetuals
Permissionless — any SPL token with liquidity can be traded
Simpler flow — SOL direct collateral, no USDC swap leg needed
Crisis resilient — decoupled from any single perpetual venue's downtime

Position data exposed

Entry price, current price, mark price
Unrealized PnL (USD), ROI percentage
Liquidation price, current LTV, liquidation LTV
Effective leverage, position size in base token
Interest accrued, daily interest cost, interest rate

$SIZE is a fair launch. No team allocation. No presale. No VC round. No locked supply with a 6-month unlock cliff we pretend doesn't exist. The token is deployed on pump.fun and the bonding curve is the distribution. There is no multisig we control, no upgrade key, no proxy contract standing between you and the tokens you hold.

Live on pump.fun

Contract Address72AyWfN3CWf3Q9h5DorqVmYnkjaAgCZjrTuB73JGpump
Venuepump.fun
Twitter@size_sol
Distribution100% fair launch
Team Allocation0%
PresaleNone
Locked SupplyNone
MultisigNone
Upgrade KeyNone
Creator WalletThe agent (autonomous)
pump.fun ↗dexscreener ↗solscan ↗twitter ↗

The split — every cycle, every fee

70% → routed through the brain into a leveraged long on the target token (PUMP in Phase 1)
30% → reserved for buyback. Once the $SIZE token is live on-chain, this bucket swaps SOL → $SIZE via Jupiter and burns the result via the SPL token burn instruction.

The 70/30 split is deliberate. 70 to leverage, because that is where the engine has to actually produce returns. 30 to burn, because that is the mechanism by which $SIZE holders get direct price support from trading activity that does not depend on the PUMP position being profitable. Even if the brain is flat for a month, the burn keeps firing every cycle and the supply keeps shrinking. Even if the PUMP side gets chopped up, the burn keeps firing. The two halves of the split are hedges for each other.

Deflationary by construction

Every cycle with non-zero fees triggers a burn (post-launch). The more the token trades, the more the supply shrinks, the more each remaining token represents a share of the engine's accumulated PUMP exposure and the engine's accumulated multi-vault revenue once Phase 2 ships. Volume is the fuel. Burn is the exhaust. The token holder sits in the driver seat, and the seat keeps getting smaller.

There is no staking. There is no LP farming. There is no yield contract that requires you to lock your tokens into a black box to earn airdrops of the next black box. Holding $SIZE is the entire participation mechanic.


You should not trust anything in this document. You should verify it. Every claim about how SIZE works can be reconciled against on-chain state and public APIs.

Live links

Agent wallet (signs all transactions)
DeMsAfZd8wUVqJEwYWKxQ74ubQ9QqBVgUDLR3CoqF9Zg
solscan.io/account/DeMsAfZd8wUVqJEwYWKxQ74ubQ9QqBVgUDLR3CoqF9Zg
Active position #1 — spot margin spot leverage on PUMP
C4YLBRkT
solscan.io/account/C4YLBRkT
Position #1 open transaction
solscan.io/tx/4YaAhySt2WCteu98sZqnpRd2w1ratSFNBjX9Y1MhKeFFJZ2F9WhZi6onyTQVyYgxuks5Bdn4si22zJddVU5ySUMJ
Active position #2 — spot margin spot leverage on PUMP
9EW9kFkd
solscan.io/account/9EW9kFkd
Position #2 open transaction
solscan.io/tx/2zKrFLn2oL5BtczkkcS3rjN9chY17osvQpsoXeaZekVce4t3Bfq9ktWQZhsPQG6EdPcKEUXdKEEzziBH5jVPNG26
Live position state (current PnL, leverage, liquidation distance)
pumpmax-site.vercel.app/api/position
Live agent decision log (every cycle, every reasoning)
pumpmax-site.vercel.app/api/logs
Current PUMP price (Pyth oracle)
pumpmax-site.vercel.app/api/pump-price

If anything in this document does not match what you can see on those endpoints right now, we screwed up and you should tell us. Building in public means the receipts are the product.


Three components, one loop. The agent claims and decides. The frontend watches and reports. spot margin, Jupiter, and Pyth are the external rails the system runs on. Redis is the shared memory. Everything else is glue.

1 · SIZE Agent (pipeline)

Node.js process running on a 15-minute cron. Handles fee claiming, splitting, brain consultation, position management, and exit monitoring. Survives RPC failures, slippage on open, and spot margin downtime. All actions logged to Redis with transaction signatures.

Claimer — monitors creator fee wallet, claims when balance exceeds threshold
Brain — OpenAI GPT-4 class entry/exit decisions with reasoning
spot margin Manager — opens and closes spot-leveraged positions
Exit Monitor — checks every position every cycle
Unwrapper — converts WSOL close proceeds back to native SOL
Buyback-Burn — Jupiter swap → SPL burn (activates with token deploy)

2 · Next.js frontend + API

Public-facing site and API layer. Serves the dashboard, docs, and real-time position data. API routes pull live data from spot margin, Pyth, and Redis on every request — nothing is cached beyond the route revalidation window.

/ — main dashboard with live position tracker, brain decisions, and chart
/docs — this page
/api/position — live position data from spot margin + Redis
/api/pump-price — current PUMP price from Pyth via Hermes
/api/logs — recent agent activity from Redis

3 · External protocols

spot margin — spot leverage protocol; borrows SOL against collateral, buys real PUMP
Jupiter — DEX aggregator for buyback swaps with MEV protection
Pyth Network — price oracle for real-time PUMP pricing via Hermes API
Upstash Redis — event logging, position registry, brain reasoning store

Multi-vault architecture (Phase 2 forward spec)

The pipeline is already designed for multi-vault. Vaults are config objects with hostToken, targetToken, splitLong, splitBuyback, and venuefields. Adding a new vault is a config operation, not a code change. The brain handles per-vault reasoning by passing the vault's target into the market context call. Per-vault stats are stored under size:vault:<mint>:* Redis keys.


All endpoints are public, return JSON, and require no authentication.

GET /api/position

Live position data from spot margin and pipeline statistics from Redis. Returned shape (real example, current state):

response.json
{
  "live": true,
  "agentWallet": "DeMsAfZd8wUVqJEwYWKxQ74ubQ9QqBVgUDLR3CoqF9Zg",
  "position": {
    "hasPosition": true,
    "count": 2,
    "positions": [{
      "address": "C4YLBRkT",
      "openTx": "4YaAhySt2WCteu98sZqnpRd2w1ratSFNBjX9Y1MhKeFFJZ2F9WhZi6onyTQVyYgxuks5Bdn4si22zJddVU5ySUMJ",
      "side": "LONG",
      "collateral": 0.129904,
      "borrowed": 0.519616,
      "entryPrice": 0.0021541872,
      "currentPrice": 0.0021749972,
      "unrealizedPnl": 0.531612,
      "roiPercent": 4.83,
      "liquidationPrice": 0.0018885509,
      "effectiveLeverage": 4.816,
      "dailyInterestCost": 0.001409,
      "currentLtv": 0.7945,
      "liquidationLtv": 0.915
    }],
    "totals": { "collateral": 0.609927, "pnl": 2.476839, "avgLeverage": 4.8165 }
  },
  "stats": {
    "cycleCount": 8,
    "longCount": 9,
    "totalLongNotional": 10.983,
    "pendingBuyback": 2.787
  }
}

GET /api/pump-price

Current PUMP price from Pyth Network via Hermes API.

response.json
{
  "price": "0.0021749972",
  "source": "pyth",
  "updatedAt": "2026-04-11T01:15:00.000Z"
}

GET /api/logs

Recent agent activity. Each log has a type (CYCLE, BRAIN, CLAIM, LONG, EXIT, BUYBACK, BURN, ERROR), a plain-English message, an optional transaction signature, and optional structured details.

response.json
{
  "logs": [
    {
      "type": "BRAIN",
      "message": "Exit decision for 7jsFgFQb: HOLD — The current ROI is +4.76%, which is between -15% and +15%. The position is not under immediate threat of liquidation, and the interest cost is manageable.",
      "details": {
        "action": "HOLD",
        "reasoning": "...",
        "urgency": "low",
        "roi": 4.76,
        "distanceToLiq": 13.157
      },
      "timestamp": 1775869036657
    }
  ],
  "stats": { "cycleCount": 8, "longCount": 9, "claimCount": 24 }
}

$SIZE involves real financial risk. This section is not a disclaimer — it is a technical assessment of what can go wrong.

Liquidation risk

At 5× leverage, a ~20% drop in PUMP price approaches liquidation
spot margin auto-sells the position when LTV exceeds the liquidation LTV (~91.5%)
The brain monitors distance to liquidation every cycle and can FULL_CLOSE preemptively
Liquidation is a SELL, not a freeze — collateral remnants flow back to the agent wallet

Borrowing cost risk

~99% APR on borrowed SOL — this is the cost of leverage
Interest deducted from collateral continuously by spot margin
In sideways or chopping markets, interest erodes the position over days
Brain weighs interest cost against ROI when deciding HOLD vs CLOSE

Brain failure modes

OpenAI API downtime → cycle skips with logged ERROR, retries next cycle
Bad reasoning → fallback to deterministic SKIP (no entry on uncertainty)
Brain output is structured JSON — schema validation rejects malformed responses

RPC and infrastructure risk

Solana RPC failures → cycle skips, no positions affected, retries next cycle
spot margin downtime → cycle skips, fees accumulate in agent wallet
Redis downtime → cycle still runs, logs cached locally, sync on recovery

Smart contract risk

spot margin contracts are permissionless and audited, but not risk-free
Jupiter swap routing carries standard DEX risk
Pipeline agent code is open-source on GitHub, single-author, and verifiable

Market risk

PUMP is a memecoin — high volatility, potential for rapid drawdowns
Leverage amplifies both gains and losses
Liquidity conditions may affect execution quality on open and close

No dates, because dates are a lie, but the sequence is not.

Phase 1 — PUMP engine · LIVE NOW

The flagship vault. SIZE creator fees route into leveraged PUMP longs via spot margin. AI brain making real entry and exit decisions. Dashboard live at pumpmax-site.vercel.app. Positions auditable on Solscan. This is the proof of concept.

Phase 2 — Multi-vault

The engine template goes public. Any memecoin project can deploy a SIZE-style vault for their own token with a permissionless flow: pick a target, pick a split, deploy. Creator fees from the host token fund longs on the target token and buybacks on the host token. One engine pattern, many memes fueled. The SIZE protocol earns a small fee on every vault, and SIZE holders get exposure to the entire fleet of engines running on the infrastructure.

Phase 3 — Hyperliquid expansion

Solana is home. Hyperliquid is the next room. We are building a Hyperliquid adapter so the SIZE brain can open perp positions on HYPE-listed memes and PUMP perps when they list. One brain, two venues. The Solana side keeps running on spot margin. The Hyperliquid side runs in parallel with its own agent, its own vault, and its own routing.

Phase 4 — Asset-backed memes

The long vision. After Hyperliquid integration, we start building the next layer: memecoins that are asset-backed by tokenized real-world assets. Oil. Gold. Commodities. Tokenized equities. Treasury bills. The SIZE engine already knows how to route SOL into long exposure. Point that same engine at a tokenized WTI crude contract or a tokenized basket of stocks, and the split changes from buy-back-the-host / long-the-meme to buy-back-the-host / long-the-real-asset.

The result is a memecoin whose price floor is reinforced by an on-chain, verifiable, automatically compounding position in something the real economy prices. A pump joke backed by a barrel of oil. A dog coin backed by a slice of the S&P. The culture stays degenerate. The collateral gets serious.

Phase 5 — Public brain

Eventually the brain itself becomes the product. A public endpoint any protocol can query to get a structured, market-context-aware leverage decision with reasoning attached. The brain as oracle. Until then it lives inside SIZE, making decisions for every vault.


Is the agent really running autonomously?
Who controls the agent wallet?
What happens if PUMP goes to zero?
Can I withdraw my fees once routed?
How is this different from just buying PUMP?
What if spot margin goes down?
What model does the brain use?
Why is the brain logged in plain English instead of just numbers?
Can other memecoins use this?
What's the minimum to participate as a holder?
$SIZE · LEVERAGED-PUMP-INFRASTRUCTURE · ALL TRANSACTIONS ON-CHAIN
home · solscan · · github · api