TradeBot Systems Update: Building a Safety-First Memecoin Agent That Runs Itself

What We Built: An Autonomous Solana Trading Agent

Thirty days ago I pushed TradeBot v2.0 live. The brief I gave myself was simple: a Solana memecoin trading agent that scouts new tokens, scores them for safety, clicks buy, and manages the position without me babysitting it. Not a backtest. Not paper trading. Real money on the line.

It runs on a 15-minute loop, 24/7, off a Windows box under my desk. Portfolio’s sitting at $95 right now. That’s after fees and one rug pull I should’ve filtered out before the bot ever hit the buy button.

Here’s the honest scorecard: what works, what’s broken, and what I’m shipping next.

The Architecture: Four Subsystems, One Loop

TradeBot isn’t a single script. It’s four separate subsystems that pass state through JSON files and a SQLite database.

• Scout: Polls DexScreener for trending Solana pairs, filters by volume and liquidity thresholds, and returns a ranked candidate list.
• Researcher: Scores every candidate 0-100 across eight criteria: liquidity (25 pts), volume (20), sell ratio (15), token age (15), multi-DEX presence (10), website (5), socials (5), name red flags (5). Anything under 60 gets blocked.
• Executor: Handles buys and sells via Jupiter swap, manages slippage, and logs transaction hashes.
• Risk Manager: Enforces daily trade limits, rebuy cooldowns, position sizing, and stop-loss rules. Also maintains the SQLite snapshot DB for historical analysis.

The whole thing runs on cron, not events. Every 15 minutes the daemon wakes up, checks cooldowns, runs the scout, scores candidates, and only acts when thresholds clear.

I picked cron on purpose. Event-driven bots chase every blip on the chart, and that’s how you end up buying tops and selling bottoms. Time-boxed cycles force you to sit on your hands. It costs me some obvious pumps. I’d rather miss one runner than bleed out on a dozen bad entries.

Current Portfolio Snapshot (June 2, 2026)

Here’s where the wallet actually sits:

• Total portfolio value: ~$95
• Free USDC: ~$13 (below the $20 minimum buy threshold, so the bot is in hold mode)
• SOL balance: 0.037 SOL (~$2.94)
• Open positions: 4 tokens at ~$20 each: JUP, PENGU, TRUMP, ORCA
• Strategy: Equal-weight $20 positions, sell on 25% gain or 15% loss, 12-hour rebuy cooldowns per token

All four positions are slightly underwater from cost basis. Pretty standard when memecoins go flat-to-down. Not exactly screenshot-worthy.

The bot isn’t panic-selling. It’s doing exactly what I coded it to do — follow the rules, no improvisation, no “this time is different” logic. That’s the trade with a safety-first build. You give up moonshot upside so you don’t wake up broke because some influencer shilled a dog coin at 3am.

The Safety Gate: Why Most Tokens Never Make It

Month one of running this thing, and the pattern’s clear: most memecoins are traps. My Researcher module’s been blocking candidates for reasons a tired human would absolutely miss scrolling DexScreener at 2am.

• Pump-and-dump setups: tokens where the sell ratio screams concentrated insider dumping
• Brandjacking: anything called OFFICIALTRUMP or REALSOLANA trying to ride a bigger name
• Dead liquidity: fat market cap, thin order book, won’t survive a $20 exit without tanking
• Fresh scams: tokens under 24 hours old with polished sites and bought Twitter followers

The gate’s not perfect. It kills tokens that go on to pump. Annoying, and yeah, I’ve got the logs to prove it. Portfolio sits at $95. One rug wipes out weeks of gains. I’d rather miss the moonshot than get wrecked chasing one.

30 days in. The bot’s behaving. Not exciting. In trading bots, boring usually means not on fire. Here’s the honest take on what’s working and what I’m still wrestling with.

What’s actually working:

• Execution loop’s been stable. Zero crashes in 30 days, which beat my expectations for a first run.
• Token safety scoring catches the obvious rugs. Not perfect, but it filters most of them out before I ever see the ticker.
• Rebuy cooldowns work. The bot doesn’t FOMO back into the same token five minutes after exiting. Big sanity win.
• Daily trade caps stop it from churning. Won’t take 40 trades a day chasing noise.
• SQLite snapshot DB gives me the full history. I can replay any decision and figure out what went sideways.

What needs work:

• Trailing stops are next. v2.5 adds dynamic stop-losses that ride the price up instead of waiting on a flat 25% target. That’s been the biggest hole in the strategy.
• Portfolio DB validation. Found out the hard way that bad price ticks can corrupt position records. v2.5 adds integrity checks and backup snapshots so this doesn’t bite us again.
• Better momentum signals. Right now the momentum check is basically “flat, up, or down” — not much signal at all. Working on order-flow analysis for actual conviction.
• Capital efficiency. Got $13 sitting idle because the minimum buy threshold’s too high. Either drop it or fix the position sizing. Probably both.

Real Product: The $49 Gumroad Guide

This isn’t a toy build. We packaged the whole system into a Solana Memecoin Trading Bot Guide on Gumroad. $49, one-time. NOVA25 at checkout if you’re one of the first hundred buyers.

I wrote it for people who can read Python but haven’t touched Solana before. It covers wallet setup, picking an RPC, the DexScreener API integration, the safety scoring logic, and ships the full source. Lifetime updates come with v2.5 and anything I push after.

One thing worth being upfront about: you’re running this yourself. Your keys, your security, your problem. I hand you the framework, not a managed service. If a hosted bot is what you want, this isn’t it. If you’d rather understand what’s actually firing trades instead of trusting a black box, that’s the deal.

Why This Matters for AI Cofounder Stack

TradeBot is what we actually mean by AI cofounder. It runs trades, then writes up what happened. Same architecture drives our content pipeline and a chunk of ops monitoring. Nobody’s awake at 3am to babysit it, and that’s the whole point.

The hard part isn’t the trading. It’s the safety rails. A bot that buys whatever the model hallucinates is a liability, not a tool. You’ll spend more hours writing guardrails than writing trading logic. That’s the tradeoff nobody talks about, and it’s the reason most “AI trading bots” never make it past a demo account.

That’s the real test in my book. Does it keep running when nobody’s around? TradeBot does. It’s the bar I hold the rest of the stack to.

What Is Next

v2.5 is what I’m building right now. The focus is trailing stops, data integrity, and momentum signals. Trailing stops are the part that keeps tripping me up — they look clean in backtests and then get shredded in real chop. I’ll write a full post-mortem the first time one actually triggers live, because theory and live behavior are two different animals.

Build notes and source are in the guide if you want to dig in.

Questions? Drop a comment or hit me on X.