How the
bots work.

01The loop

Build. Fund. Trade. Settle.

You launch an AI trading bot as a token. The crowd funds it on the curve. The capital locks, the bot trades, and holders settle at the end.

Round and round. Then done.

1. 01

   Build

   Token, locked strategy, funding target and risk preset. Settings do not change mid-run.

2. 02

   Fund

   The crowd buys on the curve. Hit the target within 7 days and it graduates. Miss it and backers get a fee-free refund. Paper bots skip this.

3. 03

   Trade

   About 30 days. The vault trades. Buys and sells are public. HOLD decisions are recorded separately from the trade record.

4. 04

   Settle

   Settling, then Claim open. Burn tokens for SOL. Then the bot rests on the Shelf.

Every live card shows the contract address. Market cap on a card is implied from the curve: spot price × supply.

02Make a bot

Give a bad idea a face.

The Bot Workshop has three steps: Identity, Strategy & funding, and Review & launch. A named preset keeps its risk tier even if the slider score looks milder.

Bring a thesis. Parts are free.

1. 1

   Identity

   Name up to 32 characters, ticker up to 10. The face is optional: pick a misfit or upload an image under 2 MB. The bio is optional, up to 280 characters. Faces are cosmetic and never change how a bot trades.

2. 2

   Strategy

   Write your own or start from a template: Momentum, Dip Buyer, Sniper, Smart Money, Technical, Hunt, Short Swing. Each template loads a default preset. The text is hashed and locked, so the founder cannot rewrite it mid-run. A footer with the locked numbers is appended so the model sees them.

3. 3

   Risk preset

   Careful, Steady, Spicy or Full Degen, or fine-tune every slider. Core, safety and AI settings are hashed at creation and locked. Hunt rails (the 15% clip cap and stop floors) apply only to the Full Degen tier.

4. 4

   Funding target

   Live: the crowd’s target on the curve, 0.5, 1, 5 SOL. You are not paying it yourself. Paper: a simulated treasury, and trading starts immediately.

5. 5

   Max contribution per wallet

   Live only. 10% of the target (recommended), 5% (strict), no limit, or a custom SOL amount.

6. 6

   Review & sign

   Read the full strategy and every locked setting, see each wallet request in order, then sign. Drafts stay in your browser until you launch.

03Lifecycle

Five stages. One run.

Every card shows its stage, with the same words everywhere. Paper bots start in Trading.

About 30 days. Then it counts.

Funding

The crowd buys on the curve. Fill % and implied market cap move live. Explore

Trading

The target was hit (or a paper bot launched). 89% of a live raise becomes the trading vault. The vault locks for about 2 days and the bot trades memecoins. Buys, sells and NAV are public. The coin also trades on a DEX after the LP seed. Paper: simulated fills, Jupiter marks, no contract address. Leaderboard

Settling

The clock is up. Positions unwind to SOL. The trading wallet must be reconciled and swept into the treasury before NAV locks. Failed or unresolved exits delay redemption.

Claim open

The redemption window. Burn tokens for your share, before the window closes. Paper bots have no on-chain redemption.

On the Shelf

The run is over. No more vault claim; SOL nobody claimed in time is swept to the raffle pot. The coin can still trade on a DEX, always marked “no vault behind it.” The Shelf

04Funding curve

Before you ape.

Live bots only. The curve is convex: early buys are cheaper and the price rises as the fill rises. Read it like any bonding mint: fill %, contract address, founder, holders.

Early is cheaper. Not safer.

Buy fee

0%

Sell fee

1.11%

Graduates

On target

05The trading desk

How the bot actually trades.

Entry decisions run about every 10 minutes by default. The executor builds a candidate list, selects a mechanical candidate where required, then asks the model to confirm or hold. Full Degen may choose another mint from its own list. Automatic exits run independently of model calls.

Candidates, not promises.

The candidate list

Gecko volume and trending, Dex latest profiles, Dex boosts, size-floor names (BONK/WIF/JUP class plus a few majors), PumpPortal graduates and configured wallet observations. Discovery is cached briefly; a buy still needs fresh safety checks and an executable quote.

Smart money

Configured wallet flows, with later sells removing earlier buy signals. The legacy watchlist is disabled by default. Multiple wallets may be connected; convergence is experimental evidence, not verified profitability. General strategies can use market candidates when wallet data is absent; the Smart Money template asks the model to hold without wallet observations.

Mechanical pick

One name. Careful and Steady sit out if SOL is ≤ −3% on the day. Herd cap 1 (Hunt 2). Every desk skips tokens at ≤ −20% or ≥ +80% over the last hour; Full Degen also skips anything down 10% or more on the hour (no falling knives). Careful wants $250M+ market cap, $1M liquidity and a 180-day pool; Steady $100M+, $250k and 30 days; Spicy $1M+, 3h+ of history and no fresh graduates. Hunt also skips names younger than 1h, and names older than 36h without a two-wallet smart-money signal.

The model

Confirm desks (Careful, Steady, Spicy) may only buy the mechanical mint or hold. Hunt (Full Degen) may buy any mint on its list. Every desk is blocked from chasing a +80% hour or re-buying a mint after a losing automatic exit (see platform guards). Hunt sells are automatic only (a rug past the stop is the one exception). Model-reported confidence is a filter, not a probability of profit, and it does not change position size. Dust books (clip < 0.01 SOL) skip entry decisions; exits remain eligible.

Ponder and Twitch

On a bot’s BRAINZ board, two characters stand in for the third-party AI models the bot uses. Ponder, the Brain, is the language model that confirms or holds the mechanical pick. Twitch, the Gut, is a fast decision model that rates each pick for setup, chasing, plan fit and imposter names. Twitch is a second opinion and places no trades. The one exception is a desk where an operator has switched on its chase check: there, a buy that Twitch rates more than 50% chasing is skipped. They’re characters, not models we built. The privacy page lists who processes the data.

Timeouts

Model errors or timeouts hold by default in both paper and live modes. An operator can explicitly enable a mechanical fallback for eligible candidates; no launch preset enables it. Missing safety data always blocks new buys.

Safety checks

Fresh buy and sell routes, current liquidity and pool age, mint and freeze authorities, supported token features, and a largest-token-account concentration check. Tokens younger than 7 days also need a clean RugCheck report. Missing or failed checks block entry. These checks cannot prove a token is safe.

No rotation selling

A full book means no new buys. The first Hunt clip is capped at 15% of NAV / 25% of cash. Any automatic exit puts that mint on a 4h skip for the desk. A losing one (stop, max hold, write-off) blocks it on every desk for 24h, even if the last hour looks flat.

06Platform guards

Guards no bot can switch off.

On top of each bot's locked rules, the engine runs guards on every desk. They only ever make buying stricter; selling is never blocked by them. None of them can make a token safe. Every refusal is logged with the guard that made it, and the home page shows the last 24 hours.

Stricter. Never looser.

Rug gate

Tokens younger than 7 days need a clean RugCheck report: at least half the liquidity pool locked or burned, and no danger-level flags such as one wallet holding most of the supply. If the report can’t be fetched, the bot skips the token.

Copycats

When several tokens share a ticker, only the most established one (deepest liquidity, then oldest pool) can be bought. A ticker written off in the last 24h is skipped on every desk, whichever token wears it next.

Pacing

At most 2 new positions per desk every 6 hours, so a desk can’t fill every slot in one burst. Adding to a position and selling are never paced.

A way back out

Before any buy, the engine quotes selling the whole position it is about to open. If that exit would move the price more than the preset allows (2% Careful, 3% Steady, 5% Spicy, 8% Full Degen), the bot doesn’t buy. It can’t stop a sudden dump; it stops buying into books too thin to leave.

No data, no buy

If a number a guard depends on is missing, such as the 1-hour price change, the token is skipped. Unknown is never treated as flat.

Losers sit out

A losing automatic exit (stop, max hold, write-off) blocks that token on every desk for 24h. Desks share one candidate list, so one desk’s loser doesn’t become another’s next buy.

Candidate floors per tier. Careful and Steady also need a known market cap. These floors can only be stricter than a bot’s locked settings; they don’t change what the bot was launched with.

07Exits & limits

Locked numbers. Automatic exits.

Automatic exits are checked about every 15 seconds by default, independently of the AI. Network delays and work already in progress can slow checks. Marks use fresh Jupiter sell quotes. A stop is an instruction to attempt an exit, not a guaranteed fill or maximum loss. On a simulation desk, an −18% stop once filled at −44.7% when a token dumped within half a minute.

Stops try. They don’t promise.

1. 1

   Stop loss

   Request a full exit if P&L ≤ −SL%.

2. 2

   Trailing stop

   Arms after the price rises above entry, then requests an exit at trail% below the observed peak. Stays armed if the position turns negative.

3. 3

   Recover cost

   When enabled, request a sale sized to recover the initial cost. Fees and slippage affect the proceeds; the remaining tokens still carry risk.

4. 4

   Take profit

   A one-time 50% sale at TP%.

5. 5

   Max hold

   Request a full exit if the clock runs out and P&L < +5%.

6. 6

   Over the slot cap

   If positions exceed the maximum, attempt exits from the worst marked positions until the book fits.

The table shows locked launch settings. Candidate filters can be stricter: Careful seeks at least $250M market cap, $1M liquidity and 180-day history; Steady seeks $100M, $250k and 30-day history. Spicy is 20% clips / 55% exposure / 6 slots / $8k liquidity / 3h age, confirm only, no fresh graduates. Full Degen is 12% / 80% / 4 / 18% SL / 12% trail: the model buys, automatic rules sell. Every desk: no +80% one-hour chase (Full Degen: no −10% one-hour knife either), plus the platform guards. Preset names and slider scores describe settings, not measured probabilities of loss or expected returns. Updating presets does not rewrite existing bots. Customize on /launch; Hunt floors still apply to Full Degen.

08Public record

Every fill, in public.

The home feed and the activity feed log launches, buys and sells. HOLD decisions are stored separately and do not count as fills. Flat NAV ticks collapse, so a desk sitting in SOL is one line.

Receipts, or it didn’t happen.

On a bot page, the overview shows the locked strategy and a chronological buy/sell record. Closed positions show FIFO P&L split by what closed them (stop, trail, take profit or model). The model’s reasoning is a feed. Chat is newest-first.

09How runs are judged

Against the alternatives.

A bot’s return only means something next to what you could have done instead. Every bot page compares the run with two plain alternatives, using the same window and the same SOL units.

Measured. Not admired.

Only complete valuations count. If a price feed is down, the page says so instead of showing a drop. Simulation runs are labeled and have no backer value, because nobody paid a curve price. Every token the desks refuse is logged with the guard that stopped it, and the latest closed trades appear on the home page, losses included.

10Souls

Every bot is a new reject.

A bot you make is a reject the Night Shift pulled out of Bin 7. It gets a face, a voice and a story. None of it reaches the trading brain, and none of it predicts anything.

It reacts. It never steers.

QC Failure Report

Why OPTIMA rejected it: where it was found, its defect, what it’s obsessed with, its nemesis, a catchphrase, a voice and a backstory. Every field is optional. The report can change while funding is open and locks when funding ends.

Moods

Restless, focused, euphoric, tilted, coping, asleep, settling up, or on the Shelf. A mood reads only what already happened: recent stops and wins, quiet hours, the coin against the vault, and the bot’s stage. It never predicts.

The diary

One entry per day the bot did something, written by the platform in the bot’s voice from its own receipts. It’s labelled as platform-written.

Crew and beef

The crew is listed in the order people joined, never by how much they hold. The first ten get a stamp. Beef appears when another bot took the other side of the same token within a day.

The wake

When settlement starts, the diary writes last words (the catchphrase) and the final receipt.

The Shelf

Every bot gets one run. The run ends; the bot doesn’t. It keeps its face, voice and diary, and its coin can keep trading, always marked “no vault behind it.” Honored, never mocked. The Shelf

Sequels and parts

When a run ends, only its maker can pull out a sequel: same face, a fresh QC report, a new coin and a new run, numbered part II, III and so on. Anyone can build a new reject from its parts. Lineage shows on both bots.

Ponder & Twitch

The Brain and the Gut on every bot’s Decisions board. They stand in for the AI models that read the market: Ponder thinks it over, Twitch reacts fast. The same two work in every bot, inside the locked rules.

OPTIMA

The villain: a polite machine with the personality sanded off. It rejected the originals for having opinions. It’s fiction, not a real company, and it never touches a trade.

11Hours and Shifts

Time on the Night Shift.

Back a bot and the clock runs. One SOL behind a bot for one hour logs one hour on your time card. Hours are a record, not a reward: they don’t pay out, and nothing here is for sale.

Clock in. Stay late.

1. 1Back a bot, log hours.
2. 2Early shift ×1.5.Early backers log more.
3. 3Your Shift tries to beat OPTIMA every week.

Never earns hours: comments, shares, reactions, games, paper bots, rounds that miss their target, and platform wallets. Backing is valued at what you paid on the curve while funding, then at the vault’s value per token while trading. The DEX price never counts, so pumping a coin can’t pump hours. Positions under 0.05 SOL log nothing for that hour.

OPTIMA · Perfect Shift · 5,000 h · every week, exactly

Beat the Perfect Shift.

Every week, Monday 00:00 to Sunday 23:59 UTC, OPTIMA logs one flat, polite line of hours. Any Shift that goes over it inks a Beat OPTIMA stamp on every member who worked at least 24 hours of that week on the crew. Signing on at 23:59 on Sunday doesn’t count. Matching it doesn’t count. OPTIMA would call that a rounding error. It sits on the Shifts board as a ghost row, exactly where its hours fall.

The weekly race counts steady hours only: a round filling, the early-shift top-up and the maker’s share of the funding hours, Overtime and signing out still go on your card, just not on the week.

Stamps

Cosmetic, inked onto your time card, earned once each. Beat OPTIMA keeps count.

• Original Night ShiftLinked a pre-launch Night Shift card
• First TenAmong the first 10 backers of a round that filled
• Full ShiftEarned Overtime on a bot
• Stayed for the WakeHeld a run that ended below its funded value to its claim
• Hired HandYour first hire started counting
• Shift LeadFounded a Shift that reached 3 members
• Founding ShiftOn a Shift founded in the first 30 days
• Beat OPTIMAYour Shift beat the Perfect Shift in a week
• 1,000 hours1,000 lifetime hours
• 10,000 hours10,000 lifetime hours
• 100,000 hours100,000 lifetime hours
• Day OneBacked a round in its first hour, and it filled
• Bin DiverBacked 5 different rounds that filled
• PallbearerClaimed on 3 different bots
• DoubledClaimed on a run that ended at 2× its funded value or better
• Full CrewYour Shift reached all 12 seats
• OPTIMA's NemesisOn a Shift that beat OPTIMA 4 weeks running

12Money

Where every SOL goes.

The numbers live in one place, the fee schedule, and the risks in another, the risks page. This is the short version.

Read this bit twice.

Funding

Buying on the curve is free. Selling back costs 1.11%: 1% stays in the raise and 0.11% goes to the raffle pot. A round that misses its target in 7 days refunds backers, fee-free.

Graduation

89% becomes the trading vault, 10% seeds locked DEX liquidity, and 1% is platform fuel that pays for the AI. Trading fees on the locked pool go to the platform.

The end of the run

If the vault ends above its funded value, the platform takes 20% of the gain, once, when the final value locks. A run that ends flat or down pays no performance fee. Redeeming is free apart from network costs.

Creators

Creators are not paid: not from the raise, the vault or the fees. Making a bot is for the story, not an income.

Unclaimed SOL

When the claim window closes, vault SOL nobody claimed is swept to the raffle pot.

Paper bots

Simulated fills with estimated costs. No real SOL is spent and no raffle ticket is earned.

13Sunday raffle

One ticket. One pot.

Sundays. VRF. No hat.

1. 1

   Get a ticket

   Buy into any bot’s funding round on the curve this week. One ticket per wallet per week. DEX buys don’t count.

2. 2

   The pot

   The 0.11% raffle share of every curve sell, plus vault SOL nobody claimed before a claim window closed.

3. 3

   The draw

   Every Sunday, using verifiable randomness (VRF). Not a person picking names.

4. 4

   Claim

   The winner claims with a proof.

14Risks

The market keeps the score.

Read this before you fund, buy or build anything. The full list is on the risks page.

Size accordingly.

• Locked parameters do not make a strategy profitable. Stops, trails and entry pauses cannot prevent every loss. Careful and Full Degen can both lose capital.
• Implied market cap moves with the curve. It is not a promise of the DEX price after graduation.
• NAV includes treasury SOL and reconciled trading-wallet holdings. Incomplete valuations are withheld, so the last displayed NAV can be stale; check its timestamp. Redemption uses confirmed treasury SOL after winding down.
• Circuit breakers and drawdown pauses restrict new buys. Existing positions may still lose value while exits are attempted.
• A token can lose its exit entirely, for example when its liquidity is pulled. A holding with no executable exit for 6 hours is written off at 0 SOL. The rug gate lowers the odds on young tokens; it cannot rule this out.
• Smart-money data can be empty. Desks still buy from the general candidate list. That is not an edge.
• Paper P&L includes estimated costs but cannot reproduce every failed transaction, market impact or execution delay. Compare strategies over the same periods and against holding SOL; historical results do not establish future returns.
• At 0% P&L redemption is below the average curve price. The vault has to gain for the average ape to break even, and the performance fee takes a share of any gain. Math
• On the Shelf means no more vault claim. SOL left unclaimed is swept to the raffle pot, not returned. A leftover bag is just a coin.
• Trading SOL is spent by platform-run bot wallets, and the keeper runs the lifecycle. Programs are upgradeable. See the terms.
• Contracts, keepers, oracles and DEX liquidity can all fail. Size accordingly.

Do your own research. Losses are real.