Build trading algorithms from natural language
Describe momentum, mean-reversion, DCA, breakout, or custom market logic and generate inspectable Python strategy code.
An AI agent that writes, validates, and deploys strategies. You bring the idea and the API key.
Auto-validation · Institutional backtests · BYOK
AI trading assistant
Finny is for traders, developers, and market builders who want to move from strategy idea to generated code, validation, and backtesting without stitching together prompts, notebooks, and scripts manually.
Describe momentum, mean-reversion, DCA, breakout, or custom market logic and generate inspectable Python strategy code.
Use historical data and metrics like Sharpe ratio, max drawdown, win rate, and profit factor to evaluate trading ideas.
Bring your own AI provider keys, review generated code, and connect broker workflows deliberately instead of using a black box.
Describe what you want in plain English. Finny researches the market, generates an inspectable Python strategy, and versions every iteration — all from your terminal.
Build, Research, or Chat — press Tab to switch. Build generates strategies, Research analyzes the market first, Chat just answers questions.
A focused subagent for every task, each with its own context and tool budget — a data extractor for OHLCV and regime classification, a news researcher for headlines via web search and Discord, and scheduled watchers that monitor deployed strategies.
Alpaca, Binance, and IBKR are live today for paper or live trading. The picker auto-switches brokers when your strategy's asset class changes.
Every algorithm gets its own versioned directory — a mission card, strategy version history, append-only logs, and per-algo data. Roll back or compare iterations.
A hidden robo-advisor mode. Answer seven inputs and Finny produces a full allocation plan — click the Finny logo in the TUI to unlock it.
Bring your own API keys. Works with Claude, OpenAI, Gemini, or local models. Your keys, your choice.
Validation + Backtesting
Any model can write trading code that looks brilliant and is secretly broken — lookahead bias, unsafe calls, overfit parameters. Finny refuses to trust its own output. Every strategy is validated, then has to survive an institutional backtest with a walk-forward overfit gate before you act on it.
Generated code is guilty until proven. Every strategy is statically checked before it's allowed to run.
engine_v2 is institutional-grade. A valid strategy still has to survive the numbers.
Install Finny, describe your strategy, and let the AI trading assistant generate a backtestable workflow. Crypto and stocks supported out of the box.
Run finny in your terminal after installation to launch the TUI.