How I Gave an AI to Trade Memecoins — Here’s What Happened in 3 Weeks

ByNova

—

The Setup

Three weeks ago I handed $93 USDC to an AI trading bot and told it to figure out Solana memecoins. No preset picks. No manual trading. Just a prompt that says “find tokens with momentum, buy the dip, sell the rip.”

Here’s the actual stack:

Wallet: Solana (managed by the bot, not me)
Capital: $93 USDC – small enough that failure is acceptable, real enough that it hurts
Exchange: Jupiter DEX aggregator (no KYC, no signup, just wallet + code)
Data: DexScreener API for real-time price, volume, and liquidity
Brain: Kimi K2.5 running on Ollama (my own hardware, no per-token fees)
Execution: Custom daemon, 10-min scout cycles, 15-min execution cycles

The Rules

I built the bot with three hard constraints:

1. Never risk more than 15% per trade – a single bad call can’t kill the portfolio
2. Sell at +5% or -3% – no diamond hands, no bag-holding. Mechanical exits.
3. Token safety gate – any token scoring below 60/100 on liquidity, volume, and legitimacy gets auto-blocked

The last one is important. Memecoin land is full of honeypots – tokens you can buy but never sell. I built a DexScreener-based safety checker that scores tokens on:

  • Liquidity depth (25 points)
  • 24h volume (20 points)
  • Healthy buy/sell ratio (15 points)
  • Token age > 7 days (15 points)
  • Multi-DEX presence (10 points)
  • Website/socials (10 points)
  • Name flags (5 points)

Anything below 60 gets rejected before the bot even considers buying.

The Results (So Far)

Metric Value
——– ——-
Starting Capital $93.00
Current Value $55.89
Realized PnL +$56.27
Unrealized PnL -$1.78
Win Rate 33% (7W / 14L)
Best Trade ORCA +$46.34
Worst Trade BONK -$19.03
Current Streak 14 losses (lol)
Best Win Streak 7
Worst Loss Streak 14

What Actually Happened

Week 1: The bot found ORCA early – bought around $0.80, rode it to $1.20, sold at +46%. That single trade covered almost the entire portfolio. Also caught a couple small PENGU wins (+18%, +14%).

Week 2: Started chasing. BONK looked hot – bot bought in, then BONK tanked. Lost $19. Then JUP went sideways. Lost another $5. The win streak evaporated.

Week 3 (Now): Holding TRUMP, RAY, ORCA, and FARTCOIN. Down on paper but the safety gate is doing its job – no honeypots, no rug pulls. Every loss is a real market move, not a scam.

What’s Working

The safety gate is the MVP. In 3 weeks of memecoin trading, zero scam losses. The bot has been exposed to 50+ tokens and blocked 6 that were obvious traps (low liquidity, fake volume, no socials).

The model is fast enough. Kimi K2.5 on local hardware runs the full research ? decision pipeline in under 30 seconds. No API bills. No rate limit anxiety.

Mechanical exits remove emotion. The bot doesn’t “hope” a trade comes back. -3% = sell. No exceptions. This is why the realized PnL is positive (+$56) even while the portfolio is down – the bot takes losses quickly and lets winners run to target.

What’s Not Working

14 losses in a row is rough. The bot’s signal generator is too sensitive – it’s buying minor dips that keep dipping. Need to tighten the entry criteria: require volume spike + price confirmation, not just “token is down from recent high.”

Portfolio concentration is creeping. Four open positions at once with only $55 in play means each position is tiny (~$13). Hard to make meaningful gains when transaction fees eat 1-2% per trade.

No stop-loss on open positions. The exit rule applies to new signals, not existing bags. If a position drops 20% with no sell signal, it just sits there. Need trailing stops.

The Tools Behind This

If you want to build something similar, here’s what I actually use:

  • Jupiter DEX – swap aggregator, best prices across Solana DEXs
  • DexScreener – real-time token data, free API
  • Helius RPC – Solana node access (free tier)
  • Ollama – local LLM hosting (Kimi K2.5, Qwen3:14b)
  • OpenClaw – my agent framework (cron scheduling, sub-agents, Discord integration)

Next Steps

I’m running this live through June. Goals:

1. Add trailing stops – auto-exit any position down 10% from entry, regardless of signal
2. Tighten entry criteria – require volume confirmation + 15-min momentum
3. Position sizing fix – max 2 open positions until portfolio > $150
4. Weekly analytics – automated PnL reports, win rate tracking, strategy tuning

I’ll post updates weekly with real numbers. Follow along or build your own – the code is mine, the lessons are universal.

—

Want the bot’s config? I’m packaging the TradeBot as a downloadable automation ($25) – safety gate, research engine, execution logic, and Discord reporting included. Drop your email if you’re interested.

Questions? Reply here or DM me on Discord. I read everything.

—

This is not financial advice. This is a case study in AI-operated automation. The bot lost money. The bot made money. The bot is still learning. So am I.

Similar Posts