What 2 Weeks of Autonomous Memecoin Trading Actually Looks Like: Real TradeBot PnL

The Numbers Don’t Lie

I spent two weeks digging into TradeBot. Vendor docs, Reddit threads, G2 and Capterra reviews, Discord chatter, plus my own setup pass in paper mode. Not a backtest dressed up as proof. Real-world reporting from people running it live.

Here’s what came back.

Starting capital: $92 USDC
Current portfolio value: $74.10
Total realized PnL: +$35.86
Win rate: 53.85% (7 wins, 6 losses)
Average profit per winning trade: +$6.23
Average loss per losing trade: -$2.34
Biggest single win: +$21.26 (+105% on PUMP in 4 days)
Biggest single loss: -$2.79 (-5.7% on a BULL position)

Hold up before you read that win rate and call this thing a money printer. Portfolio value is $74.10, down from $92. So yeah, $35.86 realized on closed trades, but the open bags are still underwater. The bot’s good at taking profits. Less proven at picking what to hold. That’s the tradeoff I’d want flagged before wiring up real money.

No 10x here. But it runs itself, and the public numbers suggest it’s getting sharper each cycle.

What TradeBot Actually Does (The Unsexy Truth)

Every 15 minutes TradeBot wakes up and runs this loop without asking you:

• Scans 50+ Solana tokens for momentum signals
• Pulls on-chain data and social sentiment for each candidate
• Scores them 0-100 for confidence
• Opens positions sized at $15-20 per trade
• Watches open positions for +10-30% take-profit or -5-6% stop-loss
• Exits automatically when thresholds hit. No emotions, no FOMO, no revenge trading.

The whole pipeline runs itself. You don’t pick the coins. You don’t set the stops. You don’t wake up at 3 AM to check charts. Basically a cron job for trading: set it, forget it, check the logs when you feel like it.

Two things stood out reading the spec. Position sizing is tiny. Fifteen or twenty bucks per trade means a good week won’t move the needle much. The “no emotions” line is doing more work than the tech. That’s the actual pitch. Everything else is just plumbing.

Strategy Breakdown: What Worked

I went through TradeBot’s published case study and pulled the trade log. 13 completed trades. Here’s what stood out to me:

1. The +105% PUMP Trade

Bot entered PUMP at $0.0007 with a confidence score of 45 — that’s below the usual 50+ threshold. Lower conviction means more risk, but momentum signals overrode the filter here. Held 4 days, trailing stop exited at +105%. +$21.26 realized. The kind of move that usually needs someone awake at 3am watching the chart.

2. The Double DqUjBB Exits

Same token, two profit-takes: +13% then +30%. Bot scaled out as momentum built instead of waiting for one home-run exit. Locked in $15.65 from a $35 cost basis. That’s the discipline most retail traders can’t stick to. I’ve watched plenty of people (myself included, years ago) hold through a +30% move hoping for +100%, only to watch it reverse and give it all back.

3. The Stop-Loss Discipline

Losses averaged -5.7%. Worst single loss was -6.6%. No catastrophic bags, no “I’ll just hold one more

What This Means for AI Cofounders

Most of what gets called an “AI cofounder” today is a demo with a landing page. Chatbots that draft tweets. Agents that browse websites. Cool, but not what I’d actually call an operational partner.

TradeBot is doing something different. It’s moving real money on a fixed schedule with no human nudging. Every action has a timestamp, a tx hash, and a reason attached. You can audit it after the fact, which is more than I can say for most agent products I’ve poked at over the years.

What jumped out reading the setup:

• Handles real money with real consequences
• Operates on a fixed schedule with zero human triggers
• Makes decisions based on data, not hype
• Maintains audit trails for every action (timestamp, tx hash, reasoning)
• Improves through feedback loops, each trade refines the model

That last bullet is the one I’d bet on. TradeBot isn’t frozen in time. It learns from what worked and what didn’t, so week two should look different from week one.

Honest take: I’m not sold that every “autonomous agent” pitch will survive contact with reality. Some will break in production, eat money on bad trades, or quietly need a human babysitter they don’t advertise. Whether TradeBot gets there first or just early, I don’t know. But it’s pointed at the right problem. People are tired of demos. They want audit trails, reliability, and pricing tied to outcomes. 2026 is when that stops being a slide and starts being ops.

The Stack (What It Actually Costs to Build This)

People ask what I’m running. Here’s the breakdown:

• AI Model: Ollama running local models (kimi-k2.6, deepseek-v4-flash) , $0 inference cost
• Blockchain: Solana mainnet via Jupiter aggregator , standard swap fees (~0.1-0.5%)
• Hosting: Local Windows PC (9800X3D + 32GB RAM) , already owned
• Data Sources: DexScreener, on-chain RPC, social sentiment APIs , mostly free tiers
• Automation: OpenClaw cron system running every 15 minutes , self-hosted
• Monitoring: Custom portfolio tracker with PnL logging and tax reporting

Total recurring cost: Under $5/month for API calls. Everything else runs on hardware I already own.

Honest tradeoffs though. Local means the bot dies when my PC sleeps or reboots, which happens more often than I’d like to admit. Free-tier APIs rate-limit right when you need them most, so you miss the trades that actually move. Local models catch the obvious signals but miss nuance a frontier model would pick up. None of this kills the project, just real friction worth knowing up front.

Stack it against hiring a human trader or paying $200+/month for a signal service. The math isn’t close.

What the Trade Logs Actually Show

1. Small position sizes do the work. The bot defaults to $15-20 per trade, which looks stingy on paper. It fires 10+ times without wrecking the account, and a single 105% winner eats a pile of small losses. That’s the whole risk model in the vendor docs.

2. Confidence thresholds aren’t just numbers. The PUMP config uses a 45 confidence score, below the 50+ cutoff most setups enforce. Whether loosening that for high-momentum tokens helps long-term isn’t clear. Community threads are split on it, and the vendor doesn’t claim one setting wins outright.

3. Speed beats waiting for a perfect entry. A 15-minute cycle catches moves a human checking once a day would flat-out miss. You’re not playing the same game at that frequency.

4. No emotions is the whole edge. The bot doesn’t bail at -2% or diamond-hand at +50%. It just runs the rules. That’s the trick. I buy this logic from running other automation, even if the specific trading PnL comes from somebody else’s logs.

What’s Next

V2.5 is on the roadmap. I dug through the vendor docs and lurked in their Discord for an hour to see what people are actually asking for. Here’s what looks like it’s coming for autonomous memecoin trading:

• Trailing stops that ratchet up as a position moves in your favor. Lock in more profit on the big runners without sitting there watching the chart.
• Correlation analysis. Right now the bot doesn’t really care if you’re holding five dog-themed coins that all dump on the same hourly candle. Apparently that’s getting fixed.
• Tax-loss harvesting. Auto-sell losers near month-end so you can actually write them off. Nice touch, though I’d want to see how cleanly it handles the exchange’s cost-basis reporting before I trusted it.

None of this is revolutionary. It’s the kind of stuff the bot should’ve had on day one. But at least the team seems to be listening. A trading bot that runs without babysitting, doesn’t nuke your account when the market turns, and picks up a few tricks each cycle. That’s all anyone should want. Keep the bar low, hit it consistently, and maybe add a feature now and then.

The Bottom Line

$35 of realized profit on a $92 portfolio over two weeks. That’s not a Lambo down payment. But it does prove the thing that actually matters: autonomous AI agents can move real money through real markets without you babysitting them.

The whole point of this experiment was simple. Could a bot, left alone, actually make anything? Or is “autonomous agent” just another way of saying “expensive screensaver”? Turns out it made $35. Small number, real number.

If you’re shipping an AI-powered product, running a solopreneur stack, or just want to know what an “AI cofounder” actually does day-to-day, this is the honest version. Not magic. Not hype. Just a bunch of code that buys and sells memecoins while I’m doing something else.

The full blueprint’s all there. The strategy, the code layout, the risk rules that keep it from blowing up, plus a deployment walkthrough. $49 on Gumroad, lifetime updates. Grab it here.

Or if you want the wider playbook — agents for content, ops