v6.0 Release Notes: Toro AI Assistant

We're pleased to announce the latest release of Toro, featuring significant improvements to search accuracy and retrieval, new ways to refine your prompts, expanded chat capabilities, and a range of interface fixes and polish across the product.

Search Accuracy & Retrieval

Toro's search engine has been substantially upgraded to return more relevant results across a wider range of queries.

• Smarter date and field filtering — Toro now does a better job of identifying and applying the right date ranges and data filters to your queries, reducing cases where relevant research was excluded by overly broad or incorrect filters.

• More resilient search strategies — When an initial search returns few or no results, Toro now automatically broadens its approach: surfacing semantically related results from a wider set, including items that match your filters even without a close text match, and retrying without restrictive filters where appropriate.

• Better title matching — Toro now better detects when users are searching by document title (exact or partial) and performs a title-specific search, surfacing strong title matches directly even when they wouldn't have appeared through the standard search path. Previously, prompts referencing a known document by name might fail even if the document existed.

• Improved effective date filtering — Date-based queries now target the correct fields, improving the accuracy of searches scoped to specific time periods.

• Keyword search for rare entities — A new parallel keyword search now runs alongside the standard vector/embedding search. When a user searches for a specific entity name (e.g. "Abivax") that exists in document body text but isn't tagged as a structured field, the keyword search catches it — addressing cases where embedding search alone couldn't reliably locate rare proper nouns.

• Correct AND/OR filter logic — Fixed an issue where the search reconciler incorrectly forced OR conditions on different field filters. For example, searching for "the latest Manager Letter for Chicago River Partners" now correctly ANDs the manager and document type filters instead of OR-ing them.

• Chronological result ordering — Logical search results (e.g. "find the last four Capital Calls") are now sorted chronologically, newest-first, even when no explicit time range filter is applied.

• Correct handling of missing date fields — The "recent" time strategy no longer silently excludes documents that don't have a custom date field populated (e.g. effective-date); documents with NULL custom date fields now correctly fall back to date_added.

• Invalid time range filters caught — Historical time range filters without valid start/end dates are now caught and filtered out before execution, preventing empty search plans.

• Author filter over-application fixed — The search planner now correctly avoids injecting the current user as an author/shared-with filter on broad queries like "identify main themes from Q1 notes" — only applying that filter when the prompt actually warrants one.

New Features

Charting in Chat Responses

Toro can now render charts in chat responses. This works similarly to existing table support — the AI can generate charts from document data using a structured response format. When a user focuses on a specific note, the response can include up to 5 charts visualizing that data, with a safeguard in place to prevent the AI from fabricating chart data in follow-up messages.

Prompts & Interaction

@ Mention Tagging in Prompts

You can now use @ in your Toro prompt to tag specific authors, lookup values, and list fields. Autocomplete suggestions appear as you type, making it easy to scope your questions to exactly the right people or categories without having to remember precise names or values. Tags already used in the current prompt are now excluded from the @ autocomplete dropdown, reducing confusion. A related issue where adding 3+ entity tokens to the chat input caused rendered chips to break — and where long messages with tokens failed to send silently — has also been fixed.

Real-Time Search Summaries

While Toro processes your query, it now displays a live summary of its search strategy (e.g. "Searching for…"). This gives you immediate feedback on how Toro has interpreted your question and what it's looking for, so you can quickly tell whether to refine your prompt or let it run.

Author Clarification

When your query references an author and Toro identifies multiple possible matches, it will now ask you to confirm which person you mean before searching. This avoids returning results for the wrong author and saves time on follow-up queries. If a user receives a disambiguation question but navigates away without answering, that state is now properly scoped and no longer interferes with the next chat.

Chat Interface & UI

Focus Mode Improvements

The interface for entering and clearing Focus mode has been refined for a cleaner, more intuitive experience. The "Clear All" button for focused documents now appears for a single document, not just 2+, matching the main webapp's workflow for easier deselection.

Cleaner Source Citations

Resolved display issues with document references in Toro responses. Cited sources for untitled notes now show "Untitled" instead of rendering as blank entries. The "Show All" button no longer appears when there's only 1 cited source, since there's nothing to expand.

Other Chat Fixes

When a user prompt contained unintelligible text, the AI would generate a garbled summary header — this is now handled gracefully. Deleted notes were previously still surfacing in Toro results due to a data sync gap; this is now handled correctly. An issue causing incorrect timestamp display when creating a new chat from a previous chat session has also been fixed.