VTX Macro’s latest update adds a new AI Screener experience designed to help users discover trade candidates before they reach the decision stage. Instead of manually scanning lists of symbols, users can configure the Screener to evaluate a ranked exchange universe, apply custom prompt context, and produce structured tradability scores with reasoning.

The Screener has its own page in the app, available from navigation and profile-aware routes. From there, users can see whether screening is running, paused, waiting for a browser owner in Client Mode, or ready for the next run. The page shows current progress, the active candidate, upcoming candidate, run timing, and recent results. Completed evaluations include tradability, reasoning, volume context, model used, latency, cost, and token information.

AI configuration now includes a dedicated Screener model panel. Users can select a Screener model separately from main and review models, tune temperature, token limits, timeout, run cadence, candidate count, loops per run, and the market universe. Screener prompts also have their own configurable system and user prompt areas, including Screener-specific context variables such as ranked volume information.

Client Mode support is a major part of this release. Browser-owned Screener runs can use the user’s local/client-side provider setup while coordinating ownership so only one tab actively runs a profile’s Screener at a time. If a tab refreshes or resumes, VTX Macro can recover active run state and continue from the correct candidate instead of duplicating work. Server Mode runs are also supported, with model usage tracked through the existing server-side billing flow.

The update also improves live feedback. Screener progress and recorded evaluations are streamed through profile-specific websocket updates, with fallback refresh behavior when websocket connectivity changes. This keeps the Screener page current while evaluations are being generated.

Several model and provider refinements are included as well. The app now handles Screener-specific model usage estimates, clearer provider error details, structured JSON output for Venice-compatible runs, and better source identity handling for model pricing and catalog display. Model selector behavior has also been tightened so Screener settings remain consistent with the rest of the AI configuration experience.

The in-app help and platform documentation have been refreshed to explain the new Screener workflow, Client Mode versus Server Mode behavior, and where Screener results appear in the product.