Building AI trading agents — curated repos, guides & backtests

Advertisement

Video Picks

Hand-picked videos on agentic AI trading — connecting AI agents to brokers, building and backtesting strategies, and honest walkthroughs of what works and what doesn't. We add new picks as we find them; suggest one.

Latest Videos

Claude's New Trading Agent Is Insane! (Tutorial)

Published 29 June 2026 · third-party video

A step-by-step tutorial on Claude's new agentic trading capabilities — connecting Claude to market data and a brokerage through MCP connectors and APIs so it can research and place trades. Look past the "insane" framing: the most useful part is seeing exactly what the wiring involves before you decide whether to hand any of it real money.

The sharpest take is in the comments — and it's effectively from Claude itself. As one viewer (@davidlowe8597) recounts: asked about trading, Claude said it can link to data and trading sites via MCPs or APIs, but warned against it — citing token costs, that it has no training in trading, and that it hallucinates. It advised that if you do connect it, to consider only its recommendations, and called any autonomous use for trading "highly dangerous." That's exactly the right frame: a model that will confidently make things up should never be left to trade unattended. If you experiment, keep a human in the loop, ring-fence a small amount and paper-trade first — see our broker-API safety notes and backtesting guide. Educational only, not financial advice.

I Wish Someone Had Shown Me How to Talk to Claude Like This Earlier (in 10 min)

Published 4 July 2026 · third-party video

Not a trading video — but a genuinely useful one if you lean on Claude for market research or to help build an agent. It breaks down three shifts — context, outcome and role — that change how Claude responds, and works one real decision four different ways so you can see each shift change the output. These are the same levers that separate a vague "analyse this stock" prompt from one that returns something you can actually act on.

The three shifts:

Context — give Claude the full situation and constraints, not just the question
Outcome — say what a good answer looks like and what you'll do with it
Role — tell it whose expertise to reason from

Pair it with our prompts for trading and Claude for trading guides.

I Built an AI Trading System With Claude + TradingView

Premiered 6 June 2026 · third-party video

An end-to-end workflow build: the creator wires together Claude Code, TradingView, Python, and AI automation tools into a complete trading pipeline from scratch. Useful for seeing how the pieces connect — charting/alerts on TradingView, logic in Python, and Claude Code as the build assistant — rather than a single black-box bot.

What it covers:

Connecting Claude Code to a TradingView workflow (charts, indicators, alerts)
Using Python to turn signals into an automated trading process
Stitching AI automation tools into a repeatable, end-to-end workflow

"A workflow" is not "an edge." Connecting Claude, TradingView and Python cleanly is the easy, satisfying part; whether the underlying strategy actually makes money after fees, slippage and drawdown is the hard part the tooling doesn't answer. TradingView alerts and webhooks can also mis-fire or lag, so treat automated execution cautiously. Educational only, not financial advice — paper-trade first and validate the strategy before risking capital; see our backtesting guide.

How to Actually BUILD a CLAUDE TRADING BOT (10 Minutes)

Premiered 30 April 2026 · third-party video

A hands-on build walkthrough: using Claude Code to assemble a fully automated trading bot and wire it to a real exchange (Hyperliquid), starting on the free testnet so you can place real positions with zero money at risk. A good, practical look at the agent build loop even if you don't have a coding background.

What it covers:

Scaffolding the bot with Claude Code — no prior coding experience assumed
Connecting to Hyperliquid and placing positions on the free testnet (paper-trading with real order mechanics)
The basic loop of an automated strategy: read data, decide, execute

Testnet ≠ live trading. Building a bot in ten minutes is the easy part; a bot that's actually profitable after fees, slippage and funding is the hard part, and a testnet gives you none of the real-money psychology or execution costs. Hyperliquid is a leveraged perpetuals exchange — moving from testnet to live means real liquidation risk, so keep it on testnet until you've validated an edge, then risk only money you can lose. Educational only, not financial advice; see our getting-started guide and broker-API safety notes.

I Tested Letting Claude Trade For A Month and Made $102k

Published 14 June 2026 · third-party video

The creator hands portfolio management to Claude for a month and reports the results — a first-person look at what a real hands-off "let the AI run it" experiment actually involves, from setup to day-to-day management to the final tally.

Treat the "$102k" headline as marketing, not a benchmark. A single one-month result — reported by the person selling access to a paid community — is anecdote, not evidence: no independent audit, no starting capital disclosed, no Sharpe or drawdown, and survivorship bias means you rarely see the months that went the other way. A month is far too short to separate skill from a lucky market. The video also promotes a paid "AI Trading Community"; that's a sales funnel, so weigh it accordingly. Educational only, not financial advice — paper-trade first, ring-fence any real capital, and see our broker-API safety notes and backtesting guide.

How To Build A Self-Improving AI Trading Agent (Insanely Cool)

Third-party video

Builds an autonomous trading agent that learns from every trade and updates its own strategy — no human in the loop. It wires the Hermes Agent framework into Claude Code with a single one-shot prompt that scaffolds the whole pipeline, from data ingestion to Railway deployment, in one session.

What it covers:

The four bars a real AI trading agent must clear — accurate data, 24/7 reliability, a defined goal with explicit success/failure conditions, and a feedback loop that actually changes its own behaviour
Why most "AI trading bots" are just prompt-in/output-out with no memory or learning
Using the scientific method to iterate — one variable at a time, baseline then improve, hypothesis then test
A first live strategy targeting Bittensor subnets with daily and 30-minute rebalances
The one-shot Claude Code prompt that installs the whole system

Important: the creator puts real capital behind a self-modifying agent — that is experimental and high-risk. A self-improving loop can overfit to recent noise as easily as it can learn, and Bittensor subnets are especially volatile. If you replicate this, paper-trade first, ring-fence a small amount, and keep monitoring. Educational only, not financial advice; see our getting-started guide and broker-API safety notes.

This Claude Fable 5-Minute Scalping Bot is SO GOOD It's Almost Illegal (Yours FREE)

Third-party video

A walkthrough of five 5-minute-timeframe scalping strategies built with Claude Fable 5, plus a demo of an AI tool the creator uses to backtest them without a TradingView account, and a Gold strategy found by running 23,000 backtests. Useful as a look at how people are using the latest models to prototype and stress-test scalping ideas.

Read the headline sceptically. Claims like "5,381%", "the US government banned it" and "almost illegal" are marketing, not verified results — treat them as such. Short-timeframe scalping is among the hardest, highest-cost styles to trade, and a strong backtest is not a live track record. Educational only, not financial advice; paper-trade first and see our backtesting guide.

Robinhood Just Let AI Trade For You: Here's How To Set It Up

Published 5 June 2026 · third-party video

A live, step-by-step tutorial that connects Claude to a Robinhood account through the new Agentic Trading beta and hands it a real "buy-the-dip" strategy with real money — a clear look at how consumer brokerages are starting to let AI agents place trades.

What it covers:

Getting into Robinhood's Agentic Trading beta, and which platforms work (Claude, ChatGPT, Cursor)
The desktop workaround to connect Claude to Robinhood using the official MCP connector
Opening and funding a separate agentic account so the AI only touches money you set aside
Giving the AI a concrete strategy: dollar-cost-average into NVDA, buy the red days, skip the green days, auto-execute
Why it still confirms before spending real money — and where agentic trading is heading

Important: letting an AI trade real money is experimental and high-risk. If you experiment, use a small, ring-fenced amount, keep human confirmation on, and treat this as education — not a recommendation. See our guide to connecting agents to broker APIs.

Building an AI Trading Agent with Python — Strands Agent & Live Coding

Published 29 Sept 2025 · third-party video

Live-codes a working AI trading agent from scratch using the Strands Agent framework and a local model — the clearest way to watch the agent loop (plan → act → review) actually come together.

Reinforcement Learning Trading Bot in Python — Train an AI Agent on Forex (EUR/USD)

Published 12 Dec 2025 · third-party video

Trains a reinforcement-learning agent on historical EUR/USD data and tests it — a hands-on companion to our FinRL build guide and a good intro to the learn-a-policy approach.

I Built an AI Trading Bot in 10 Minutes with Claude Code (Python + Interactive Brokers)

Published 9 Apr 2026 · third-party video

Uses Claude Code in VS Code to scaffold a Python trading bot wired to Interactive Brokers — a fast look at AI-assisted agent building. Treat the "10 minutes / profitable" framing sceptically and paper-trade everything first.

Build AI Trading Agents with OpenAI's Agent Builder — Multi-Agent Workflow

Third-party video

Walks through a multi-agent trading workflow in OpenAI's Agent Builder — useful for seeing how specialised agents hand off to each other, the same pattern behind TradingAgents.

Build a Python Trading Bot for Algorithmic Trading Using AI — Full Tutorial

Published 28 Apr 2025 · third-party video

A complete, beginner-friendly walkthrough of a real-time algorithmic trading bot in Python — a solid foundation for the plumbing (data, orders, scheduling) before you add an AI decision layer.

Advertisement

Build along: the repos behind these videos

Want to build what these videos demo instead of just watching? These are the open-source repositories they lean on — the CLI to scaffold a bot, the exchange and broker APIs to connect it, and a ready-made bot to build on. All are third-party and open source; the GitHub links are not affiliate links. For agent frameworks and backtesters, see our full Best Repos list.

GitHub · Anthropic

The agentic terminal CLI several of these videos use to scaffold a trading bot from a prompt.

GitHub · Exchange SDK

hyperliquid-python-sdk

Official Python SDK for Hyperliquid — the exchange in the Claude bot build. Has a free testnet for zero-risk practice.

GitHub · Connectivity

One unified API to 100+ crypto exchanges — the connectivity layer most bots are built on.

GitHub · Trading bot

A mature, open-source crypto trading bot with built-in backtesting — a solid base to build on rather than starting from scratch.

GitHub · Broker API

Python API for Interactive Brokers (the community successor to ib_insync) — pairs with the IBKR build video.

GitHub · Agent framework

The model-agnostic agent framework from the Strands live-coding walkthrough.

Backtesting Walkthroughs

An agent is only as trustworthy as its backtest. These walkthroughs cover the exact frameworks on our Best Repos page — pair them with our Backtesting AI Agents guide (which has runnable code).

Backtest Your Trading Strategy IN MINUTES with VectorBT

Published 2024 · third-party video

A quick, practical VectorBT walkthrough — load data, define entries and exits, and run a backtest fast. VectorBT is the speedy, vectorized engine built for parameter sweeps.

Backtesting.py — Full Course in Python

Third-party video

A full course on Backtesting.py — the lightweight library we use in our backtesting guide. The friendliest place to run your very first backtest.

High-Performance Backtesting and Trading with NautilusTrader (Part 1)

Third-party video

A deep, technical session on NautilusTrader — the production-grade, event-driven platform where the same strategy code backtests and trades live. For when you mean to deploy.

Introduction to Backtrader — Backtesting Strategies in Python

Third-party video

A clean introduction to Backtrader's event-driven model. Heads up: Backtrader's development has largely stalled, but it's still an excellent way to learn how event-driven backtesting works.

Backtrader vs VectorBT vs Zipline vs NautilusTrader: Which Should You Use?

Published 2025 · third-party video

Can't decide which engine to start with? This recent comparison weighs Backtrader, VectorBT, Zipline and NautilusTrader head-to-head — the perfect companion to our Best Repos backtesting picks.

More on Agentic Trading

Build Your First AI Trading Agent

LLM choice, framework and paper-trading setup.

Backtesting an LLM Trading Agent

Avoiding look-ahead bias and overfitting.

Connecting Agents to Broker APIs

Paper vs live, order types and safety.

AI Platforms for Trading

Using Claude, ChatGPT & Gemini.

Videos are from third-party creators and shared for education only; inclusion isn't endorsement and nothing here is financial advice. Featuring a video does not imply any affiliation with the creator or broker. Some links may be affiliate links; see our disclosure.