ChangelogChangelog

The predictions agent now uses Tavily for deep web research when evaluating markets, with DuckDuckGo as a fallback. Richer context leads to better-calibrated probability estimates.

Replaced the internal `is_active` concept with an explicit `enabled` flag for broker connections. Stock-only brokers are now filtered from crypto and futures trading automatically.

Added a twikit-based Twitter client for monitoring market-moving tweets and sentiment signals. The agent can now factor in real-time social chatter alongside traditional data sources.

Predictions and user settings panels now require an explicit Save button press rather than auto-saving on every change, preventing accidental configuration updates.

Added a `/llm-rankings` endpoint backed by live API data rather than static rankings, giving an up-to-date leaderboard of model performance across tiers.

Integrated Resend webhook handlers to track email delivery status (sent, opened, bounced) and feed that data back into notification reliability monitoring.

Updated to compliance agreement v3.0 with seven new SEC/FINRA disclaimer clauses and an electronic signature requirement during onboarding.

Enabled PostHog's cookieless mode to improve privacy for users who opt out of tracking, removing third-party cookie dependencies from the analytics pipeline.

Every change to broker keys, user settings, and compliance agreements is now written to an immutable audit log with timestamp, user ID, and diff — meeting financial compliance requirements.

Changed compliance enforcement from fail-open to fail-closed: if the compliance check errors out, trading is blocked rather than permitted. Safer default for regulated environments.

Your subscription tier is now shown during AI conversations, giving the model context about your available features and model access.

Improved position reconciliation across multiple connected brokers for more accurate portfolio views.

Reduced minimum liquidity thresholds for predictions, allowing more markets to be analyzed and scored.

Updated the AI models available on the free tier to provide better quality analysis out of the box.

Added detailed architecture documentation to the web UI explaining how the autonomous trade loop works internally.

Upgraded from Gemini Flash-Lite 1.5 to 2.0 across all services for faster inference and improved analysis quality.

New profile dropdown in the navigation for quick access to account settings and dashboard.

AI engine caching now uses time-based keys to prevent stale analysis from persisting across market sessions.

Upgraded to Gemini 3.1 Flash-Lite with reordered AI provider priority for optimal cost-performance balance.

AI model routing now supports per-provider tier overrides, allowing fine-grained control over which models serve each subscription tier.

New marketing service capabilities for analyzing SEO performance and discovering content opportunities.

Added position size calculator, risk/reward calculator, and options profit calculator to the web tools section.

Programmatic SEO pages comparing supported brokers with dynamic routes, RSS feed, and sitemap integration.

Crypto and futures trades now track detailed exit reasons (target hit, stop loss, regime shift, EOD flatten) for post-trade analytics.

Crypto screening now fetches data in parallel, cutting scan cycle time significantly.

Futures trades now automatically submit initial protective stop orders on entry, with correct limit price sign handling for Alpaca and Schwab.

Prediction agent now respects per-user settings, allowing individual customization of prediction parameters.

AI conviction scoring now processes candidates in batches, reducing latency and token usage per scan cycle.

Refactored 40+ conversation action handlers from a 400-line if/elif chain into a clean registry pattern for better maintainability.

New daily loss recovery system that automatically adjusts position sizing and risk parameters after significant drawdowns.

Dynamic exit engine now scales trailing stop distances based on real-time volatility, tightening in calm markets and widening in volatile ones.

Futures trading now re-resolves broker connections at execution time, rejecting inactive broker records to prevent failed orders.

Crypto data providers now use task-local context managers for broker scoping, preventing data leaks between concurrent user sessions.

Options strategies, legs, and journal entries now persist their broker association for accurate tracking across sessions.

Added granular user preference mappings for 50+ alert types, giving users fine-grained control over which notifications they receive.

New interactive product tour that launches after onboarding, with the ability to reset and replay it from settings.

Switched prediction market discovery to event-based fetching to avoid parlay floods, with improved volume/liquidity filtering.