FAQ
What FiClaw is, how to try it, how it differs from a general AI assistant, how backtesting and optimization work, and where its boundaries are.
FiClaw is a quant strategy research workstation. You describe a strategy idea, and it helps structure the rules, generate code, evaluate factors, run a historical backtest and compare parameter settings. The desktop app includes roles such as quant researcher and risk manager.
Yes. FiClaw Desktop is available for macOS Apple Silicon, Windows x64 and Linux x64. After installing you can log in and reach the main workspace to try the built-in financial agents, cloud models and the credit-based membership center.
A general assistant can discuss strategy logic. FiClaw also connects that work to runnable Python, historical backtests, result analysis and another reviewable test. The distinction is the surrounding workflow, not a promise that every research decision is automatic.
Yes. FiClaw connects to real A-share historical market data through QuantAPI (supporting stock pools such as the CSI 300), with backtest windows spanning 10+ years, and outputs professional metrics including Sharpe ratio, annualized return, max drawdown, win rate and turnover.
FiClaw extracts tunable parameters from the strategy code (such as momentum window, number of holdings and weight caps), generates a candidate grid around a ±50% range, and backtests each combination. It also supports walk-forward checks so you can review whether the result depends on one narrow setting.
No. FiClaw does not promise investment returns and does not replace the professional judgment, compliance review or risk-control responsibility of licensed institutions. It is meant to support research, analysis, review and pre-execution checks, not to bypass human confirmation and move money on its own.
No. FiClaw works better as a collaborator and workflow amplifier: it reduces repetitive work, improves handoffs and strengthens review. Investment decisions, risk conclusions, going live, execution and external outputs should still keep human review and internal confirmation.
No. The realistic path is to start with one key step, give the system a clear role there, and expand only when the result is useful. This keeps the scope controllable and makes it easier to judge what is genuinely effective.
FiClaw works best when it enters from a real bottleneck and expands once the boundaries are clear. Let’s discuss a pilot scope.