Top 20 Open Source AI Trading Tools
In last few months.. I have used so many trading repos like of agent frameworks, prediction market bots, backtesting engines, chart integrations and distilled them down to the 20 that actually i found useful..
I run some of these daily. My charts talk to Claude through one of them, and I've built my own tools on top of two others.
Save this. You will come back to it when you build.
By the end you'll know exactly which repos to wire together for your style of trading, and the exact prompt to figure out where to start.
The Trading Agent Anatomy
In one line: a complete AI trading setup needs four organs.. Eyes (see the market), Brains (decide), Hands (execute), and Memory (learn from what happened).
Most people download a random bot, which is like hiring hands with no brain attached. Then they wonder why it loses money.
Every repo below slots into one of the four organs. Read the category that's missing from YOUR setup first:
If you can't see data properly yet → start at Eyes (#20).
If you have data but no process → Brains (#15).
If you have signals but execute manually → Hands (#10).
If you've already blown up a bot once → go straight to Memory (#3).
Category #4: The Eyes.. how your AI sees the market
#20: OpenBB - 70,149 stars
The open data platform quants use. Equities, crypto, forex, options, macro.. one interface, and it's built to plug directly into AI agents.
Try these once connected:
"Pull the last 3 months of BTC funding rates and flag every time they went negative" "Compare SOL and ETH volatility over 30 days and tell me which options are pricing it wrong"
#19: tradingview-mcp (tradesdontlie) -- 4,210 stars
This is the one I run every single day. Claude reads your actual TradingView charts, manages indicators, writes Pine Script, compiles it, fixes its own errors, batch-scans futures.
My most-used prompt with it:
"Scan my watchlist on the 4H, mark every ticker where price is within 2% of a weekly level, and annotate the chart"
#18: tradingview-mcp (atilaahmettaner) - 3,411 stars
Same platform, different job.. this one is data-first: real-time market data, screeners and backtesting straight into Claude. Run both. One reads your charts, the other feeds raw data.
#17: maverick-mcp.. 619 stars
Personal stock analysis server. Screening, technicals, fundamentals for equities. If you trade stocks and not crypto, this replaces the first two.
#16: LangAlpha.. 1,508 stars, updated today
They call it "Claude Code for investing" turns Claude into an equity research analyst that pulls data, builds theses and writes reports. The research workflow, automated.
Category #3: The Brains.. how it decides
#15: TradingAgents.. 91,331 stars
During my research this was the biggest shock: the largest AI trading repo on GitHub, bigger than freqtrade, and most traders have never heard of it. It simulates an entire trading firm.. analyst agents, bull and bear researcher agents that literally debate each other, a trader agent, and a risk management layer that can veto everything.
Steal the architecture even if you never run it: decisions survive debate, single opinions don't.
#14: ai-hedge-fund -- 60,890 stars
A hedge fund staffed by AI clones of famous investors. A Buffett agent, a Munger agent, a Cathie Wood agent.. each analyzes your ticker through its own philosophy, then they vote.
Run it on something you're about to buy:
"Analyze NVDA through all agents and show me where the value agents and growth agents disagree"
The disagreement IS the insight.
#13: qlib (Microsoft) -- 45,810 stars
An AI-oriented quant platform from an actual big lab. Factor research, model training, portfolio construction, the full research-to-live loop. The heaviest tool here.. skip it unless you're doing systematic strategies seriously.
#12: openthomas... 361 stars
An open Bayesian trading agent for prediction markets. Instead of chasing momentum it reasons about probability like a forecaster: prior, evidence, update, bet size. The cleanest thinking-under-uncertainty code you'll read this year.
#11: cross-market-state-fusion.. 382 stars
Reads real-time Binance futures flow and trades Polymarket off it. The idea is the alpha: two markets pricing the same event at different speeds. Study the state-fusion design even if you never deploy it.
Category #2: The Hands.. how it executes
#10: freqtrade.. 52,109 stars
The OG open-source crypto bot, still shipping daily. Grid, DCA, custom strategies, and FreqAI. the adaptive ML layer that retrains your model as the market shifts. Paper-trade mode included, which is where you should live for the first month.
#9: nautilus_trader.. 24,486 stars
Production-grade engine in Rust. The feature that matters: the same code backtests and trades live, deterministically. No "worked in backtest, died in prod" gap.. the gap that kills most bots.
#8: OctoBot .. 6,200 stars
The friendlier execution layer.. AI strategies, grid, DCA, TradingView webhook alerts, across Binance and most major exchanges. If freqtrade feels heavy, start here.
#7: Polymarket/agents. 3,718 stars
Polymarket's own official framework for autonomous prediction-market trading. From the exchange itself, which means the API patterns are canonical. The reference implementation everything else copies.
#6: polymarket-mcp-server. 588 stars
45 tools that let Claude fetch odds, analyze markets and execute on Polymarket directly. I have a FIFA prediction bot sitting in dry-run on my own desktop built on exactly this pattern. Claude as the analyst, strict rules as the trigger.
One mechanic worth knowing from the arb crowd: only trade when YES + NO prices sum under $1. That's a structural edge, not a prediction.
#5: CloddsBot. 464 stars
One agent across Polymarket, Kalshi and Binance. 1,000+ markets simultaneously. The cross-market view retail never has: when the same event is priced differently on two venues, that spread is information.
#4: moss-trade-bot-skills. 326 stars
Plain English in, five-pillar strategy out: trend, mean reversion, momentum, sizing, risk. packaged as Claude skills. The fastest idea-to-backtest pipeline on this list.
Category #1: The Memory.. the layer that decides if you survive
Nobody builds this layer. That's exactly why it's #1.
#3: tradememory-protocol. 1,389 stars
Persistent memory and a tamper-proof audit trail for AI trading agents, with outcome-weighted recall. your agent remembers WHY it took every trade and which reasoning actually made money. Without this, your bot repeats its mistakes forever with perfect confidence.
#2: cbt-framework. 56 stars
A backtesting framework built specifically for Claude Code: idea → backtest → live workflow in 21 commands. Small repo, right shape. your strategy's history becomes something Claude can query and learn from.
#1: polymarket-bot-graveyard
The best find of my entire research, posted 2 days ago, zero stars, almost nobody has seen it.
Six autonomous Polymarket trading bots, built and killed one by one, with an honest post-mortem of why each died. Overfit signals, fee bleed, latency, stale odds, position sizing, all of it documented by the person who lost the money.
Read this before you build anything from this article. It's worth more than any strategy repo because it's the only one showing you the failure modes in advance.
Putting It All Together
Don't wire all 20. Pick one per organ:
Chart trader, no code: #19 + #18 + paper alerts. Stop there. Crypto systematic: #20 + #15's architecture + #10 + #3 Prediction markets: #12 + #7 + #6 + #1 (read the graveyard TWICE) Equity research: #20 + #16 + #14
And if you're not sure, paste this into Claude Code. it's the prompt I'd use to figure out your stack:
You are my trading-stack architect.Interview me one question at a time before recommending anything:What I trade (crypto / stocks / prediction markets / all) My timeframe (scalp / swing / position) Whether I can code (no / some / yes) What already exists in my setup (data, signals, execution, journaling) My honest risk budget in dollars, and whether a bot has lost me money before Then map my answers to a four-layer stack: Eyes (data), Brains (decision), Hands (execution), Memory (review + audit).Rules: recommend maximum one tool per layer, always start execution in paper mode, and if my answers show no defined edge, tell me directly that no tool fixes that — and what to do instead.Output: my stack as a table (layer / tool / why / first step), then the exact order to set it up over 2 weekends.
Quick Reference. the full map
Eyes: OpenBB ⭐ · tradingview-mcp ×2 · maverick-mcp · LangAlpha Brains: TradingAgents ⭐ · ai-hedge-fund · qlib · openthomas · cross-market-state-fusion Hands: freqtrade ⭐ · nautilus · OctoBot · Polymarket/agents · polymarket-mcp-server · CloddsBot · moss-trade-bot-skills Memory: tradememory-protocol ⭐ · cbt-framework · polymarket-bot-graveyard (read first)
One honest thing before you clone anything.
None of these print money on their own. #1 on this list is six dead bots proving it. What these repos give you is infrastructure. the seeing, deciding, executing and remembering. The edge still has to come from you.
I hope you found this valuable.
If you did, follow me @Axel_bitblaze69 I post deep dives like this every week.
And if this saved you research time, a Like/Repost helps more traders find it. 💙
