REPLYOPSAI BUILD LOG
LAB JOURNAL · PAPER RESEARCH

Building V6 With Freqtrade and FreqAI

V6 was designed to reduce the amount of custom trading infrastructure we had to invent while preserving control over the research process. Freqtrade handles the trading framework; FreqAI supplies the ML workflow; our experiment defines what is trained, traded and measured.

Current experiment architecture

OKX DATA→FEATURES→FREQAI→SIGNAL→PAPER→METRICS
SETTINGV6
MarketsBTC, ETH, SOL
Timeframe5 minutes
Training window30 days
RetrainingEvery 6 hours
ExecutionPAPER / dry-run
Max open trades3

Why Freqtrade/FreqAI

The move was not an assumption that FreqAI is profitable. It was an engineering decision: use a mature framework for execution, state and ML integration so the experiment can focus more clearly on features, targets, model behavior and forward results.

Infrastructure boundary

The trading VPS contains execution state and credentials. The public website is separate. A small exporter converts experiment results into sanitized metrics, which are pushed over HTTPS to a read-only public data path.

TRADING VPS→SANITIZE→HTTPS INGEST→PUBLIC RESULTS

No exchange API key, raw production configuration, position quantity or bot-control endpoint is required by the website.

What we measure

Closed/open trades, realized PAPER PnL, win/loss counts, expectancy, profit factor, closed-PnL drawdown, a cumulative realized-PnL path and recent sanitized trades. The sample size remains visible because attractive early metrics can be statistically weak.