Changelog: 2026-06-21
Chat is now far more reliable, preventing “completed” sessions without an answer and eliminating a crash in the reasoning dropdown.
- Improved chat reliability so conversations don’t end without a final response, with clearer error handling when something goes wrong.
- Fixed a production chat UI crash related to the reasoning/thinking dropdown.
- Expanded research automations with a new company screening capability and more dependable filings/insider data results.
- Streamlined connector permissions and improved sign-in privacy by reducing requested access to only what’s needed.
- Strengthened billing and event processing to avoid duplicate handling and reduce edge-case inconsistencies.
- Made sandboxed sessions more resilient, reducing blank reloads and improving stability for long-running tasks.
- Chat sessions now avoid persisting placeholder “final” messages and will surface a visible error if the system can’t produce a real answer, rather than silently completing.
- Increased headroom for complex chats so longer reasoning and larger outputs are less likely to be cut off mid-response.
- The reasoning dropdown in chat was updated to prevent a crash and improve overall UI robustness.
- Added a new company screener that helps rank and filter opportunities using multiple signals, while improving how insider transactions and regulatory filings are presented when data is incomplete or delayed.
- Connector authorization was simplified by limiting requested permissions to a smaller, more standard set.
- Billing and webhook processing was hardened with better deduplication and race-resistant handling, improving reliability during high activity.
- Sandbox improvements include restoring session continuity after reloads, better durability of session history, and more robust execution for long-running headless workflows.