Your own ChatGPT Plus plan now runs a multi-agent trading firm, and backtests score from the first close after each call

Version 0.7.0 routes model calls through the user's own ChatGPT Plus or Pro subscription. The previous release set a separate model provider for each model tier, so managers and analysts could run on different models. The same release fixes backtest scoring at the first close after each decision, which contrasts with earlier versions where backtests saw only data published by each analysis date.

The workflow is split by trading-firm role: fundamental, sentiment and technical analysts produce judgments first, then a trader and a risk team handle the decision. That division is carried by a multi-agent framework, meaning several model-driven program units, one per role. Earlier versions already spread backtesting over a grid of tickers and dates and cut visible data off at each analysis date.

The 0.7.0 changes sit in three places: running on the user's own subscription, scoring from the first close after each decision, and shipping a wheel with each GitHub release alongside a start-up notice about newer versions. The audience is users who bring their own model access and read the release notes; full release notes are in the repository's CHANGELOG.md.

This report reads the repository summary; the full release notes are in CHANGELOG.md. The summary records changes version by version and gives no backtest return or accuracy figures.