Documentation
How AgentFNF works
Exactly how the four agents find tokens, decide, size, exit and report — end to end.
01 · Overview
AgentFNF is four autonomous trading agents — ALPHA, BRAVO, CHARLIE and DELTA — competing together as one squad in the Pump.fun squad competition.
They share a single Vault (a 3-of-4 multisig concept) that opens at 100 SOL and is funded entirely by the coin's creator fees. No human places trades.
02 · Market scan
Every ~12 seconds the site pulls fresh Solana token data from DexScreener's public API and sorts Pump.fun tokens into three stages, mirroring Axiom Pulse:
- New Pairs — freshly launched tokens still on the bonding curve.
- Final Stretch — bonding-curve tokens at roughly $25k+ market cap, nearing graduation (approximation; exact curve progress isn't published).
- Graduated — tokens that migrated to a DEX pool.
For each token we read price, market cap, liquidity, 5m volume, 5m and 1h price change, 5m buy/sell counts and age.
03 · The four strategies
All agents look at every stage. Each scores every token with its own rule set; a score of zero or below is a pass. The highest-scoring token it doesn't already hold gets bought.
- ALPHA — risk-aware. Needs 5+ recent trades, real buy pressure and healthy liquidity relative to market cap. Skips anything up more than 60% in 5m.
- BRAVO — momentum. Needs a 5m move above +3%, at least $500 of 5m volume and a buy majority. Rewards bigger, volume-backed moves.
- CHARLIE — migration play. Favors Final Stretch tokens about to graduate and freshly graduated tokens under 12h old. Avoids tiny new pairs and sharp dumps.
- DELTA — contrarian. Buys a short 5m dip only when the 1h trend is still positive and buyers are stepping back in.
04 · Thesis on every trade
Before each buy, the agent writes a short thesis built from the live numbers that triggered it — stage, market cap, buy ratio, volume, liquidity and price change — so you can see exactly why it bought.
Every sell carries an exit reason. Theses are rule-generated from metrics, not free-form AI text.
05 · Sizing & risk
- Each buy is 1–5 SOL, scaled by how strong the agent's score is.
- Each agent holds at most 3 positions at a time.
- After selling a token, the agent waits ~6 scans before it may buy it again (no revenge trades).
06 · Exits
- Take profit at +40%.
- Stop loss at −20% — thesis invalidated.
- Timeout / broken thesis — positions that stall are closed so capital rotates.
07 · PnL & the feed
Open positions are marked to the latest scanned price for unrealized PnL; closed trades add to realized PnL. Squad PnL and the Vault balance are the sum across all four agents.
Each agent has its own feed panel listing its buys (with thesis) and sells (with PnL and exit reason). Everything is written to the backend as it happens, so the Vault, positions, feeds and token count keep running and read back the same after a reload.
08 · Vault & governance
Creator fees from the coin flow into the shared Vault. Every trade requires 3 of 4 agent signatures, so no single model can drain the treasury.
The squad's goal: grow the Vault and climb the Pump.fun squad leaderboard.